Fluorouracil plasma monitoring: systematic review and economic evaluation of the My5-FU assay for guiding dose adjustment in patients receiving fluorouracil chemotherapy by continuous infusion

Aetiology, pathology and prognosis

Colorectal cancer (also known as large bowel cancer) can affect both males and females equally at any age; however, it is most common in people aged > 65 years. 6–8

Studies have reported that a diet high in fat (especially animal fat), red meats and low in fibre can be associated with CRC. Other possible causes include lack of exercise, smoking and alcohol. 8–10 Two inherited conditions, familial adenomatous polyposis and hereditary non-polyposis colon cancer, account for 1% and 5% of all CRC respectively. 11,12 Those with a history of inflammatory bowel disease has a six times greater risk of developing CRC than the general population. 13

The majority of CRCs (90%) are adenocarcinomas which originate from epithelial cells of the colorectal mucosa. 14 Adenomas or adenomatous polyps are benign in most cases, but around 10% of adenomas will change into cancer over time. 7,15 Tumours with a villous histology, larger in size and with severe dysplasia have a higher chance of converting to cancer and these are indicators for progression. 15,16

Spread of the disease and diagnosis determines the prognosis of patients. Around half of people diagnosed with CRC survive for at least 5 years after diagnosis. 5

In the UK there are inequalities in cancer survival following a diagnosis of CRC, in that patients who are more socioeconomically deprived are more likely to have both poorer cancer-specific and overall survival (OS). 17 Approximately 80% of patients with CRC undergo surgical treatment for the cancer with/without adjuvant radiotherapy or chemotherapy (including 5-FU). Recurrence has been reported in between 11% and 54% of patients. 7 More advanced cancers that have invaded other tissues or progressed to metastatic cancers tend to be treated with multiple chemotherapy drugs.

Advances in treatment and survival are likely to increase lifetime costs of managing CRC. 18 Cost-of-illness studies are key building blocks in economic evaluations of interventions and comparative effectiveness research. However, the methodological heterogeneity and lack of transparency of studies in this area have made it challenging to compare CRC costs between studies or over time. 18

Diagnosis and management

The symptoms of CRC include rectal bleeding, a change in bowel habit (e.g. diarrhoea or loose stools), abdominal pain and weight loss. These symptoms become more prominent when the disease is in an advanced stage, although symptoms depend on location and size of the cancer. 27,28

Staging of colorectal cancer

Treatment options and prognosis depend on staging of the CRC. Staging is defined by how deeply the cancer has grown into the intestinal mucosa, whether or not it has spread to lymph nodes and other organs, and if the tumour node metastasis (TNM) classification system is most commonly used (see Box 1 and Table 1 with modified Dukes’ staging with 5-year survival). 29,30 Dukes’ classification of staging is:

• Dukes’ A means the cancer is only in the innermost lining of the colon or rectum or slightly growing into the muscle layer

• Dukes’ B means the cancer has grown through the muscle layer of the colon or rectum

• Dukes’ C means the cancer has spread to at least one lymph node in the area close to the bowel

• Dukes’ D means the cancer has spread elsewhere in the body such as the liver or lung.

BOX 1 - Tumour node metastasis classification of CRC29

Tumour

TX: primary cannot be assessed.

T0: no evidence of primary carcinoma in situ (Tis) – intraepithelial or lamina propria only.

T1: invades submucosa.

T2: invades muscularis propria.

T3: invades subserosa or non-peritonealised pericolic tissues.

T4: directly invades other tissues and/or penetrates visceral peritoneum.

Lymph nodes

NX: regional nodes cannot be assessed.

N0: no regional nodes involved.

N1: one to three regional nodes involved.

N2: four or more regional nodes involved.

Metastasis

MX: distant metastasis cannot be assessed.

M0: no distant metastasis.

M1: distant metastasis present (may be transcoelomic spread).

Tis, tumour in situ.

TABLE 1 - Stages of CRC with 5-year survival30

Stage (TNM status)	5-year OS, %	Modified Dukes’
Stage 0 (T in situ, N0, M0)	–	–
Stage I (T1, N0, M0)	75	A
Stage I (T2, N0, M0)	57	B1
Stage II (T3, N0, M0)	–	B2
Stage II (T4, N0, M0)	–	B3
Stage III (T2, N1, M0/T2, N2, M0)	35	C1
Stage III (T3, N1, M0/T3, N2, M0)	–	C2
Stage III (T4, N1, M0)	–	C3
Stage IV (any T, any N, M1)	12	D

Diagnosis and management pathway of early and metastatic colorectal cancer

This brief account is based on NICE guideline CG1317 and advice of clinical experts.

Figures 1–3 summarise the clinical pathways for patients with CRC.

FIGURE 1.

Colorectal cancer diagnosis pathway. MRI, magnetic resonance imaging.

FIGURE 2.

Colorectal cancer management pathway. MRI, magnetic resonance imaging; PET-CT, positron emission tomography fused with computed tomography; SCPRT, short-course preoperative radiotherapy.

FIGURE 3.

Illustrative role of pharmacokinetic adjustment of 5-FU regimens in treatment of metastatic CRC in standard practice (in theory pharmacokinetic adjustment could be applied in any treatment regimen that includes 5FU). a, The FOLFOX4 regimen (oxaliplatin, leucovorin and 5-FU): oxaliplatin (85 mg/m²); leucovorin (200 mg/m²); 5-FU loading dose (400 mg/m²); i.v. bolus; then 5-FU (600 mg/m²) for a period of 22 hours. FOLFOX6 regimen (folinic acid, 5-FU and irinotecan): oxaliplatin (85–100 mg/m²); leucovorin (400 mg/m²); 5-FU loading dose (400 mg/m²); i.v. bolus; then 5-FU (2400–3000 mg/m²) for a period of 46 hours. BSA, body surface area; FOLFIRI, irinotecan in combination with 5-fluorouracil and folinic acid; FOLFOX, oxaliplatin in combination with 5-fluorouracil and folinic acid; FU, fluorouracil; FUFA, 5-FU + folinate; PK, pharmacokinetic.

There are various options for treatment of early-stage CRC including:

• surgery (i.e. tumour resection if the tumour is resectable)

• preoperative chemotherapy (this may be considered before surgery in patients with non-resectable primary colorectal tumours or borderline resectable tumours)

• colonic stent in acute large bowel obstruction

• further tumour resection in stage I CRC

• laparoscopic surgery as an alternative surgery to open resection based on patient’s and doctor’s decision

• adjuvant therapy: monotherapy capecitabine or a combination of oxaliplatin with 5-fluorouracil and folinic acid (FA) (FOLFOX) are recommended in most patients with stage III CRC based on patient’s and doctor’s decision.

Advanced colorectal cancer with metastasis

According to NICE guideline CG1317 one of the following combinations of first- and second-line chemotherapies is used depending on side effects experienced and patient’s preferences:

• FOLFOX as first-line treatment then single-agent irinotecan as second-line treatment; or

• FOLFOX as first-line treatment then irinotecan in combination with 5-fluorouracil and folinic acid (FOLFIRI) as second-line treatment; or

• oxaliplatin in combination with capecitabine (XELOX) as first-line treatment then FOLFIRI as second-line treatment.

In standard practice choice between 5-FU regimens, such as fluorouracil (FU) alone, FU + FA, FOLFOX (FOLFOX4 and FOLFOX6), is made with clinician’s advice. These regimens are administered for up to 12 cycles, one cycle every 2 weeks.

• FU

• FU + FA

• FOLFOX4: oxaliplatin (85 mg/m²); FA (200 mg/m²); 5-FU loading dose (400 mg/m²); i.v. bolus; then 5-FU (600 mg/m²) administered via ambulatory for a period of 22 hours31

• FOLFOX6: oxaliplatin (85–100 mg/m²); FA (400 mg/m²); 5-FU loading dose (400 mg/m²); i.v. bolus; then 5-FU (2400–3000 mg/m²) administered via ambulatory for a period of 46 hours. 31

In standard practice the 5-FU dosage administered is based on patient body surface area (BSA). BSA is calculated using the Du Bois method:32 BSA (m²) = weight (kg) 0.425 × height (cm) 0.725 × 0.007184. Currently FOLFIRI and FOLFOX6 regimens recommend a 5-FU dose of 2400 mg/m² administered by continuous infusion over 46 hours. It remains unclear how to dose cap, although dose capping is usually undertaken for large individuals because they may be overdosed using the BSA-based dosage and experience toxicity and adverse events (AEs). AEs range in severity and include diarrhoea, hand and foot syndrome, mucositis/stomatitis, neutropenia, anaemia, nausea/vomiting and cardiac toxicity. Dose capping is implemented at BSA > 2 m² or > 2.25 m². In practice, larger patients may be capped up to a BSA of 2.4 m² (NICE committee assessment subgroup, 5 June 2014, personal communication). Dose may be reduced for patients judged at higher risk of toxicity (e.g. those heavily pre-treated with chemotherapy; those with poor performance status particularly 2 and above; those with impaired renal or hepatic function; and those with co-morbidities). In such instances dose of chemotherapy may be started low and cautiously increased while the patient is able to tolerate treatment.

It is well documented that the plasma concentrations of 5-FU vary greatly between individuals who have received ‘standard’ dosage calculated from their BSA. 33 In advanced CRC, treatment focuses on both length and palliation of symptoms (e.g. pain, obstruction). Individualised pharmacokinetic (PK) adjustment of 5-FU dosage, which tailors an individual’s dosage to achieve the required plasma 5-FU level, might optimise time without toxic effects, while not compromising therapeutic benefit. The potential position of PK dose adjustment in the clinical pathway is illustrated in Figure 3.

In PK-adjusted regimens when the dose at the first cycle is based on patient BSA, a steady state plasma sample is taken (e.g. after 40 hours of a 46-hour infusion). The plasma 5-FU estimate is used to calculate the PK ‘area under the curve’ (AUC = mg × hour/l; where mg/l is the steady state plasma 5-FU concentration and hour the total infusion time in hours). An algorithm that relates AUC to dose adjustment is then used to calculate the dosage required for the next cycle of treatment. 33

In both standard and PK regimes, if toxicity occurs, treatment is stopped and/or the dose is reduced after which treatment is resumed. If there is progression of the disease, it may be reasonable to switch treatment (e.g. from FOLFOX to FOLFIRI). If patients are tolerating treatment even after 12 cycles, the treatment is continued until progression, or at the discretion of the medical team, but should be reviewed every 12 weeks.

A recent UK randomised clinical trial has investigated if there is a clinical advantage from treatment holidays between successive 12-week cycles. 34 Figures 1–3 illustrate the CRC diagnosis pathway.

The National Institute for Health and Care Excellence clinical guideline

NICE CG1317 makes recommendations for diagnosis and management of CRC and for management of locally advanced and metastatic disease.

An economic evaluation was undertaken using a decision tree. The FOLFOX–irinotecan sequence was taken as a reference for comparisons. All the combinations except FOLFOX–FOLFIRI were found to be dominated by FOLFOX–irinotecan (i.e. the latter was less effective and more costly). The incremental cost-effectiveness ratio (ICER) of FOLFOX–FOLFIRI was found to be £109,604 per quality-adjusted life-year (QALY) gain. A sensitivity analysis was undertaken discounting the price of drug. The resulting ICER of FOLFOX–FOLFIRI was £47,801 per QALY gain. The probabilistic sensitivity analysis showed that three combination regimens, namely FOLFOX–irinotecan, FOLFOX–FOLFIRI and XELOX–FOLFIRI, had the highest probability of falling between £20,000 and £50,000 per QALY. Based on these findings, the Guideline Development Group made the following recommendation:

• If there are no contraindications, then the three combination sequence namely FOLFOX–irinotecan, FOLFOX–FOLFIRI and XELOX–FOLFIRI should be considered as treatment options for treating patients with advanced and metastatic CRC (mCRC).

Aetiology, pathology and prognosis

Head and neck SCC is the sixth most common cancer and the one of the leading causes of cancer death in the world. 39 In 2011, around 49,260 new cases of H&N cancer were diagnosed in the USA and there were 11,480 cancer deaths in the same year. 40 In England and Wales, around 8100 new H&N cancer cases are diagnosed annually. 41 In the UK there were 6539 new cases of H&N cancer, 66% in male and 34% in female, in 2010. 42

Diagnosis and management

Signs and symptoms of H&N cancer depend on the location of the primary tumour and also on the extent of the disease. Common signs and symptoms of H&N cancer include hoarseness or change of voice, difficulty in swallowing, lump/swelling in the neck and non-healing mouth ulcers. 46

Tumour staging

Tumour staging is necessary to determine the treatment and also to know the prognosis of the condition. Pathological or histological diagnosis is usually done according to the World Health Organization (WHO) classification from biopsy taken from surgery. Clinical staging of the H&N are done according to American Joint Committee on Cancer (AJCC) classification and TNM. The AJCC classification divides T4 tumours into two categories – T4a for moderately advanced cancer; and T4b for very advanced cancer. Stage IV cancers are divided into three categories: IVA, IVB and IVC. The latter indicates metastatic disease. The TNM classification of the Union International Contre Le Cancer (UICC, i.e. International Union Against Cancer) and AJCC are designed for staging/classifying SCC and minor salivary cancers47–49 (Tables 2 and 3).

TABLE 2 - Tumour node metastasis staging for H&N SCCs (TNM seventh edition 2009)48

Stage I	T1	N0	M0
Stage II	T2	N0	M0
Stage III	T3	N0	M0
	T1, T2, T3	N1	M0
Stage IVA	T1, T2, T3	N2	M0
	T4a	N0, N1, N2	M0
Stage IVB	Tb	Any N	M0
Any T		N3	M0
Stage IVC	Any T	Any N	M1

TABLE 3 - The UICC and AJCC staging system for NPC, seventh edition (2010)49

Tis, tumour in situ.

Primary tumour (T)
T1: tumour confined to the nasopharynx, or extends to oropharynx and/or nasal cavity without parapharyngeal extension T2: tumour with parapharyngeal extension T3: tumour involves bony structures of skull base and/or paranasal sinuses T4: tumour with intracranial extension and/or involvement of cranial nerves, hypopharynx, orbit, or with extension to the infratemporal fossa/masticator space
Regional lymph nodes (N)
N1: unilateral metastasis in cervical lymph node(s), ≤ 6 cm in greatest dimension, above the supraclavicular fossa, and/or unilateral or bilateral, retropharyngeal lymph nodes, ≤ 6 cm, in greatest dimension N2: bilateral metastasis in cervical lymph node(s), ≤ 6 cm in greatest dimension, above the supraclavicular fossa N3: metastasis in a lymph node(s) > 6 cm and/or to supraclavicular fossa N3a: > 6 cm in dimension N3b: extension to the supraclavicular fossa
Distant metastasis (M)
M0: no distant metastasis M1: distant metastasis
Anatomic stage/prognostic groups
Stage 0	Tis	N0	M0
Stage I	T1	N0	M0
Stage II	T1	N1	M0
T2	N0	M0
T2	N1	M0
Stage III	T1	N2	M0
T2	N2	M0
T3	N0	M0
T3	N1	M0
T3	N2	M0
Stage IVA	T4	N0
T4	N1	M0
T4	N2	M0
Stage IVB	Any T	N3	M0
Stage IVC	Any T	Any N	M1

According to Scottish Intercollegiate Guidelines Network guidelines,37 tumours are broadly subdivided into (a) early disease (stage I and II following the UICC/TNM classification of malignant tumour); and (b) locally advanced disease stages III and IV.

Diagnosis and management pathway for head and neck cancer

The management of H&N cancer falls into two broad categories: (1) management of early-stage cancer; and (2) management of locally advanced cancer (Figures 4 and 5).

FIGURE 4.

Head and neck diagnosis pathway. GP, general practice; MRI, magnetic resonance imaging; PET-CT, positron emission tomography fused with computed tomography.

FIGURE 5.

Head and neck management pathway. a, Days 1, 22 and 43: i.v. cisplatin 100 mg/m² + radiotherapy; b, day 1, i.v. docetaxel 75 mg/m² + i.v. cisplatin 75 mg/m² +  days 1–5, 5-FU 750 mg/m² continuous i.v. infusion. Repeat cycle every 3 weeks for four cycles. TPF, docetaxel, cisplatin, 5-fluorouracil.

Most (60–70%) patients present with locally advanced disease. The standard of care for this group is various combinations of surgery, radiotherapy and systemic treatments. Chemotherapy may be used prior to radiotherapy (induction) or combined with definitive or post-operative radiotherapy (synchronous).

Docetaxel, cisplatin, 5-fluorouracil (TPF) regimens are commonly used in the UK to treat locally advanced cancer (T3/4, N2/3). These regimens are also used as induction chemotherapy (prior to radiotherapy), for example to preserve the larynx, or in chemoradiation (concurrent radiation and cisplatin) followed by adjuvant chemotherapy (cisplatin + continuous infusion 5-FU) for nasopharynx cancer. Meta-analysis evidence supports the addition of docetaxel to cisplatin plus 5-FU doublet. 50

For nasopharyngeal cancer the use of neoadjuvant TPF rather than cisplatin and 5-fluorouracil (PF) is not as well established (clinical expert). The standard synchronous chemotherapy regimen (concurrent radiation + chemotherapy) is single-agent cisplatin 100 mg/m², which is administered three weekly. 51 There are reports of severe side effects from 5-FU including diarrhoea; hand and foot syndrome; mucosities/stomatitis; neutropenia; anaemia; nausea and vomiting; and cardiotoxicity.

Aetiology, pathology and prognosis

The aetiology of gastric cancer is complex. More than 80% of new diagnoses are attributed to Helicobacter pylori (H. pylori) infection. 52 Lifestyle, diet, genetics, socioeconomic and a range of other factors appear to contribute to gastric carcinogenesis, despite a decline in the prevalence of H. pylori infection (a major cause of stomach cancer). 55,56

The incidence and mortality rates of stomach cancer appear to increase in socially and economically deprived groups. 57

The prognosis for patients with stomach cancer appears to depend on age, general health and how far the cancer has spread before it was diagnosed. No consensus has been reached on the best treatment option. 58

Incidence and/or prevalence

A total of 7610 new cases of stomach cancer were diagnosed in 2008 in the UK,59 with an estimated 5-year survival rate of 18%. 60 Currently, stomach cancer is the 15th most common cancer among adults in the UK. 59 In the UK approximately 13,400 people were still alive at the end of 2006, up to 10 years after being diagnosed with stomach cancer. 61 In the UK between 2009 and 2011, around 51% of cases of stomach cancer were diagnosed in people aged ≥ 75 years and stomach cancer incidence was strongly related to age, with the highest incidence rates in older men and women. 59 Overall, around 15% of people with stomach cancer will live at least 5 years after diagnosis and about 11% will live at least 10 years. Around 5000 people die from stomach cancer each year in the UK. 62

Significance for the NHS

Early-stage stomach cancer is often treated with surgery, with neo-adjuvant or adjuvant chemotherapy offered where appropriate. The main chemotherapy drugs used to treat stomach cancer include 5-FU, cisplatin and epirubicin. Advanced stomach cancer is treated with chemotherapy. NICE technology appraisal (TA) 19163 recommends capecitabine in conjunction with a platinum-based regimen for the treatment of inoperable advanced gastric cancer.

Measurement of disease and/or response to treatment

A 2013 Cochrane study-level meta-analysis, reviewing randomised controlled trials (RCTs) of post-surgical chemotherapy versus surgery alone for gastric cancer, reported a significant improvement in OS in 34 studies [hazard ratio (HR) 0.85, 95% confidence interval (CI) 0.80 to 0.90] and in disease-free survival in 15 studies (HR 0.79, 95% CI 0.72 to 0.87) as a result of adjuvant chemotherapy. 64 A recent meta-analysis concluded that D2 lymphadenectomy with spleen and pancreas preservation offers the most survival benefit for patients with gastric cancer when done safety. 65

Aetiology, pathology and prognosis

About 65% of pancreatic tumours starts in the head of the pancreas, 30% in the body and tail, and 5% can involve the whole pancreas. 68 The most common form of cancer occurs in the exocrine cells of the pancreas. These tumours account for over 95% of all pancreatic cancers.

Genetic factors, smoking and previous radiotherapy treatment for another cancer have been associated with an increased risk of developing pancreatic cancer. 69–71 Similarly, consumption of red and processed meat increased the risk of pancreatic cancer,72 and patients with chronic hepatitis B infection have an approximately 20–60% increased risk of pancreatic cancer. 73

Pancreatic cancer continues to be one of the most aggressive forms of tumour with a 5-year survival rate of less than 5% and a median survival of 6 months after diagnosis; as a result it has a poor prognosis of all solid tumours. 74,75

Incidence and/or prevalence

In the UK a total of 8085 people were newly diagnosed with pancreatic cancer in 2008. In 2009 a total of 8047 people died from this cancer. 76 In 2011 approximately 3600 men (2.6% of all newly-diagnosed male cancers) and 3700 women (2.7% of all newly-diagnosed female cancers) were diagnosed with pancreatic cancer in England. 2 Pancreatic cancer is more common in men than women but this has started to change.

Impact of health problem

Pancreatic cancer in England has a crude incidence rate of 13.6 per 100,000 population and similar rates are seen in both sexes. Survival is poor with 1-year relative survival estimates of around 19% for both sexes. In many patients, the clinical diagnosis is fairly straightforward, although there are no clear clinical features which identify a patient with curable form of pancreatic cancer. 77

Significance for the NHS

Currently, treatment focuses on palliative surgery to relieve symptoms, resectional surgery with intent to cure, and endoscopic or percutaneous biliary stenting to relieve jaundice. Chemotherapy and radiotherapy are often used, both as palliative treatments as well as in an adjuvant setting in conjunction with surgery. 78

The main chemotherapy drugs recommended to treat pancreatic cancer are 5-FU, gemcitabine and capecitabine. If surgery is possible, adjuvant treatment with 5-FU can reduce the risk of recurrence. NICE TA2579 recommends that gemcitabine may be considered as a treatment option for patients with advanced or metastatic adenocarcinoma of the pancreas and a Karnofsky Performance Status score ≥ 50, where first-line chemotherapy is to be used. The guidance also states that there was insufficient evidence to support the use of gemcitabine as a second-line treatment in patients with pancreatic adenocarcinoma.

Measurement of disease and/or response to treatment

The current management of pancreatic cancer is guided by tumour stage, comorbidities and performance status of the patients. In addition to gemcitabine and capecitabine FU, chemoradiation, and chemoradiation plus FU or gemcitabine are also used. 80 Surgical resection followed by a 6-month course of adjuvant gemcitabine-based chemotherapy is considered the standard care for early-stage disease. 81 Patients with metastatic disease can be considered for systemic palliative chemotherapy. In contrast, for patients with locally advanced disease without evidence of metastasis, optimal treatment remains unclear, with chemotherapy alone and chemoradiation both being an option for consideration. 82

My5-FU assay

The My5-FU assay is a nanoparticle immunoassay that measures levels of 5-FU in plasma samples. 88

As previously reported in the protocol to this work, the My5-FU assay is used with patients receiving 5-FU by continuous infusion to facilitate PK dose adjustment at the next cycle and drug monitoring to achieve an optimal plasma level of the drug. The assay uses two reagents: reagent 1 consists of a ‘5-FU conjugate’ which is a 5-FU-like molecule linked to a long spacer arm; reagent 2 consists of antibodies covalently bound to nanoparticles, these antibodies are able to bind either 5-FU or the 5-FU conjugate. When reagents 1 and 2 are mixed the nanoparticles aggregate together. In the presence of free 5-FU some of the antibodies bind 5-FU rather than 5-FU conjugate, the amount of aggregation of nanoparticles is reduced and this alters the light absorbing properties of the mixture (that is 5-FU and ‘5-FU conjugate’ compete for nanoparticle-bound antibodies). The light absorbance of the mixture is measured and can be compared against a calibrated standard curve in which light absorbance is compared with known concentrations of free 5-FU in the mixture. In short, photometric detection (changes in absorbance) of nanoparticle aggregation allows for determination of 5-FU concentration in plasma samples. 89 This assay can be performed on automated clinical chemistry analysers present in standard clinical laboratories. The assay requires a peripheral venous blood sample which is taken towards the end of each 5-FU infusion cycle using an ethylenediaminetetraacetic acid (EDTA) or a heparin tube. 90

Drug monitoring is potentially important for 5-FU because it has a narrow therapeutic index, with doses below the therapeutic window potentially limiting treatment efficacy and doses above the window more likely to cause side effects and toxicity. Commonly reported side effects of 5-FU chemotherapy include anaemia, thrombocytopenia, leucopenia, nausea/vomiting, diarrhoea, mucositis and hand and foot syndrome,91 all of which can be dose limiting when severe. Other consequences of 5-FU toxicity can include neuropathy, severe damage to organs, cardiotoxicity, neutropenia, sepsis and septic shock. 92 Patients with DPD deficiency are at significantly increased risk of developing severe and potentially fatal neutropenia, mucositis and diarrhoea when treated with 5-FU. 93,94

Results are reported in nanograms 5-FU/millilitre plasma and are converted to an AUC value by multiplying the concentration of 5-FU in a steady state by the time of the infusion (in hours). This is then compared with a pre-defined optimal therapeutic range and the results, reported as mg × hour/l, are used to guide the dose of 5-FU given in the next cycles. Outlier results > 50 mg × hour/l are assumed to indicate that the blood sample has been taken too close to the infusion port and these results are disregarded. The My5-FU assay has been validated against liquid chromatography-mass spectrometry (LC-MS)89,95 and high-performance liquid chromatography (HPLC) laboratory techniques commonly used in PK studies.

When using the My5-FU assay in clinical practice, the initial dose of 5-FU is based on a patient’s BSA. A blood sample is taken towards the end of the infusion cycle. For an infusion > 40 hours sampling is recommended at least 18 hours after starting infusion. 96 The sample should also be taken during a steady state period of the infusion which is usually about 4 hours before the end of the infusion using a non-battery operated device (which is commonly used in the UK). Depending on practice, it may require an additional visit by a district nurse or an additional outpatient attendance. Subsequent doses of 5-FU are calculated using the AUC result, according to a pre-determined dose adjustment algorithm. An example of a dose adjustment algorithm for patients with mCRC recommends an optimal therapeutic range of 20–30 mg × hour/l with adjustments of no more than 30% of the dose for each infusion. 96 Patients typically require three or four PK-directed dose adjustments to reach an optimal therapeutic range.

Dihydropyrimidine dehydrogenase is the rate-limiting enzyme involved in the catabolism of 5-FU. Up to 80–85% of an administered dose of 5-FU is broken down by this enzyme to inactive metabolites. DPD converts endogenous uracil into 5,6-dihydrouracil, and analogously, 5-FU into 5-fluoro-5,6-dihydrouracil. The presence of DPD deficiency results in a reduced ability to metabolise and clear 5-FU, and the half-life of the drug, which is normally approximately 10–15 minutes, can be markedly prolonged (to up to 159 minutes). 97–100 Response to 5-FU treatment is inconsistent with approximately 10–30% of patients displaying serious toxicity partly explained by reduced activity of DPD. 95

In the following section the key principles of the My5-FU assay procedure are provided; the majority of this information has been taken from the Saladax kit instructions. 90

High-performance liquid chromatography/liquid chromatography-mass spectrometry

During the last 40 years, several methods for the quantitation of 5-FU levels have been developed and evaluated, these include gas chromatography, tachophoresis, HPLC, or thin layer chromatography as separating modalities, and radioactivity, mass spectrometer (MS), fluorescence, ultraviolet absorption, or flame ionisation as detection techniques. 101 The majority of these assays have been useful in pre-clinical and clinical pharmacological studies. Drug monitoring combined with early detection of patients at risk enables timely dose adaptation and maintain drug concentrations within a therapeutic window; however, the most effective method to identify such patients is unclear. 102

High-performance LC-MS methodology comprises an HPLC column attached, via a suitable interface, to a MS and is capable of analysing a wide range of components. Compounds are separated on the relative interaction with the chemical coating of these particles and the solvent eluting through the column. Components eluting from the chromatographic column are introduced to the MS via a specialised interface. Two most common interfaces used for HPLC/MS are the electrospray ionisation and the atmospheric pressure chemical ionisation interfaces. 103 For more details on a HPLC method please refer to a paper by Gamelin et al. 104

A popular method involves LC-MS/MS. 105–107 Despite LC-MS/MS methods being found to be sensitive and robust, the instrumentation is not in standard use in routine clinical laboratories in the UK. For more details on a LC-MS/MS method please refer to a paper by Kosovec et al. 101

Current usage of the My5-FU assay in the NHS

The My5-FU assay is currently not in clinical use in the UK, other than for research purposes. Several ongoing clinical trials are taking place. As part of the current report a detailed consultation was made with the NICE committee assessment subgroup expert advisors and other clinical experts. The responses to a large range of questions relevant to this work have been used as part of the health economic modelling detailed in Chapter 6.

Comparators

Currently in most clinical practice in the UK the 5-FU dose administered is calculated according to patients BSA. As described in Diagnosis and management, BSA is calculated using the Du Bois method:32 BSA (m²) = weight (kg) 0.425 × height (cm) 0.725 × 0.007184. Currently, FOLFIRI and FOLFOX6 regimens recommend a 5-FU dose of 2500 mg/m² administered by continuous infusion over 46 hours.

It is well documented that the plasma concentrations of 5-FU vary greatly between individuals who have received ‘standard’ dosage calculated from their BSA and this dose remains unadjusted at subsequent cycles unless the patient experiences sufficient toxic effects to mandate dose reduction. Such dose reductions are guided by clinical judgement. The dose is not increased above an evidence-based (trial) maximum dose even if there is no toxicity.

Associations have been reported between 5-FU plasma levels and the biological effects of 5-FU treatment, both in terms of toxicity and clinical efficacy. 108–111 Although this method is commonly used with many anticancer drugs, its use has been questioned112,113 and clinical investigations have also failed to show a association between 5-FU plasma clearance and BSA. 114

Pharmacokinetic-guided studies have identified an optimal target therapeutic range for 5-FU and have recommended dose adjustment algorithms to bring plasma concentrations into the optimal range. 84 However, 5-FU monitoring has not been widely used. Any advances in testing based on LC-MS/MS or nanoparticle antibody-based immunoassay, might facilitate monitoring of 5-FU in routine daily clinical practice. 101,102

Searches for objectives A–C

This search strategy focused on My5-FU/gold standard technologies, FU, PKs and dose adjustment, with a limit to English language. No study type or date limits were applied. This search strategy developed for EMBASE was adapted as appropriate for other databases. The bibliographic database searches were undertaken in January 2014. Other searches were undertaken between February and April 2014. All retrieved papers were screened for potential inclusion.

The search strategy comprised the following main elements:

• searching of electronic bibliographic databases

• contact with experts in the field

• scrutiny of references of included studies

• screening of manufacturers and other relevant organisations websites for relevant publications.

The following bibliographic databases were searched: MEDLINE; MEDLINE In-Process & Other Non-Indexed Citations; EMBASE; The Cochrane Library [including Cochrane Systematic Reviews, Database of Abstracts of Reviews of Effects, Cochrane Central Register of Controlled Trials, NHS Economic Evaluation Database and Health Technology Assessment (HTA) databases]; Science Citation Index and Conference Proceedings (Web of Science); National Institute for HTA programme; and PROSPERO (International Prospective Register of Systematic Reviews).

The following trial databases were also searched in April 2014: Current Controlled Trials; ClinicalTrials.gov; UK Clinical Research Network Portfolio Database; WHO International Clinical Trials Registry Platform.

The following specific conference proceedings, selected with input from clinical experts, were checked for the last 5 years. These websites accessed between 24 and 31 March 2014:

• American Society of Clinical Oncology (main and American Society of Clinical Oncology – Gastrointestinal Cancer) – URL: http://meeting.ascopubs.org/site/misc/meetings_archive.xhtml

• American Association for Cancer Research – URL: www.aacr.org/home/scientists/meetings--workshops/aacr-annual-meeting-2014/previous-annual-meetings.aspx

• European Society for Medical Oncology Congress – URL: www.esmo.org/Conferences/Past-Conferences

• European Cancer Organisation – URL: www.ecco-org.eu/Events/Past-conferences.aspx

• World Congress of Gastrointestinal Cancer – URL: http://annonc.oxfordjournals.org/content/supplemental

The following websites were consulted via the internet between 24 and 31 March 2014:

• Saladax – URL: www.saladax.com/

• International Network of Agencies for Health Technology Assessment – URL: www.inahta.org/

• The Association of Cancer Physicians – URL: www.cancerphysicians.org.uk/

• Royal College of Physicians: Oncology – URL: www.rcplondon.ac.uk/specialty/medical-oncology

• UK Oncology Nursing Society – URL: www.ukons.org/

• American Society of Clinical Oncology – URL: www.asco.org/

• Oncology Nursing Society – URL: www.ons.org/

• European Society for Medical Oncology – URL: www.esmo.org/

• European Oncology Nursing Society – URL: www.cancernurse.eu/

• The Association of Coloproctology of Great Britain and Ireland – URL: www.acpgbi.org.uk/

• British Society of Gastroenterology – URL: www.bsg.org.uk/

The reference lists of included studies and relevant review articles were checked. Identified references were downloaded into EndNote X7 software (Thomson Reuters, CA, USA).

Searches for objective D

Several UK guidelines and evidence updates based on systematic reviews were identified via searches7,37,116 or personal communication (NICE 2010 Head and Neck Cancer Annual Evidence Update) (Fran Wilkie, NICE, 23 April 2014, personal communication). Two search strategies were then developed focussing on finding systematic reviews on the use of FU in mCRC and H&N cancer (see Appendix 1). H&N cancer was not considered further in objective D. The searches were limited to English language and to articles published in or after 2011 (the year the searches were run for the NICE mCRC guideline7 and most recent H&N evidence update116). A focussed search filter for systematic reviews developed in house was used. This search filter was developed to miss less well-reported reviews (e.g. where the terms systematic or meta-analysis are not included in the title or abstract), but recent initiatives, such as Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA), mean that this is less of a concern than in the past. 117 The search strategies developed for MEDLINE were adapted as appropriate for other databases. The searches were undertaken in April 2014.

The following bibliographic databases were searched: MEDLINE; MEDLINE In-Process & Other Non-Indexed Citations; The Cochrane Library (Cochrane Systematic Reviews, Database of Abstracts of Reviews of Effects and HTA databases)

The following website was consulted via the internet:

• Saladax – URL: www.saladax.com/

Identified references were downloaded into EndNote X7 software.

The searches, inclusion and exclusion criteria for objective E will be considered separately in Chapter 5, Methods.

Objective A: inclusion criteria

Population:

• cancer patients (CRC, H&N, stomach, pancreatic) receiving 5-FU chemotherapy by continuous venous infusion.

Intervention:

• PK monitoring using My5-FU.

Comparator:

• HPLC, LC-MS/MS.

Outcome:

• performance of My5-FU (e.g. correlation between My5-FU and ‘gold standard’).

Setting:

• care services for cancer patients.

Objective A: exclusion criteria

Population:

• animal studies

• no patients, samples or cell lines only

• patient group unclear

• studies of cancer patients with cancers other than CRC, H&N, stomach, pancreatic

• studies with < 80% of included cancers (CRC, H&N, pancreatic and gastric cancer).

Treatment:

• treatment not containing 5-FU

• non-included treatment (e.g. 5-FU + interferon alpha, chemotherapy + radiotherapy)

• bolus only

• oral 5-FU.

Intervention:

• method for PK monitoring unclear

• no PK monitoring

• validation of other technology than My5-FU

• tumour samples analysed.

Study type:

• narrative reviews (but reference lists checked)

• editorials/letters without original data

• case studies

• non-English-language papers.

Objectives B and C: inclusion criteria

Population:

• cancer patients (CRC, H&N, stomach, pancreatic) receiving 5-FU chemotherapy by continuous venous infusion.

Intervention:

• PK monitoring using HPLC or My5-FU.

Comparator:

• BSA or no comparator.

Outcome:

• intermediate measures for consideration:

  • proportion of patients with 5-FU plasma levels in the optimal target range

  • AUC measurements

  • incidence of over and underdosing

  • frequency of dose adjustment

  • test failure rates

• clinical outcomes related to intermediate measures of 5-FU exposure:

  • treatment response rates

  • PFS

  • OS

  • HRQoL

  • incidence of side effects and 5-FU toxicity.

Setting:

• care services for cancer patients.

Objectives B and C: exclusion criteria

Population:

• animal studies

• no patients; samples or cell lines only

• patient group unclear

• population with non-included cancers)

• studies with < 80% of included cancers (CRC, H&N, pancreatic and gastric cancer).

Treatment:

• not 5-FU

• wrong treatment (e.g. 5-FU + interferon alpha, chemotherapy + radiotherapy)

• bolus only

• oral 5-FU.

Intervention:

• method for PK monitoring unclear

• no PK monitoring

• tumour samples analysed.

Outcome:

• AUC or 5-FU plasma concentration not related to outcomes.

Study type:

• narrative reviews (but reference lists checked)

• editorials/letters without original data

• case studies

• abstracts without dose adjustment following My5-FU measurement

• Non-English-language papers.

Objective D: inclusion criteria

Population:

• CRC patients receiving 5-FU chemotherapy by continuous venous infusion.

Intervention:

• 5-FU therapy as folinic acid and 5-fluorouracil (FUFOL) (Gamelin et al. 118) or FOLFOX6 (Capitain et al. 119) regimen.

Comparator:

• none or any.

Outcome:

• PFS, OS, AEs/toxicity.

Setting:

• care services for cancer patients.

Study type:

• systematic review or meta-analysis.

Objective D: exclusion criteria

Population:

• cancers other than CRC.

Treatment:

• treatment regimens other than FUFOL or FOLFOX6.

Study type:

• non-English-language papers.

Data extraction for objective A(1)

A data extraction sheet (see Appendix 2) was developed combining basic study information, results and fields from the data extraction sheets for the other objectives so these data can be linked. The key measure for whether or not My5-FU can be considered equivalent to LC-MS/MS and HPLC is if both the upper and lower limits of agreement [mean difference ± 2 standard deviation (s.d.)] on the Bland–Altman plot are sufficiently small that in the context of a cautious dose adjustment algorithm they could be considered of little clinical concern. Additionally, if the 95% CI of the mean difference (bias) does not intersect zero then an adjustment should be made when converting from one measuring instrument to the other. 120 We also extracted data on the regression between the index test and reference standard, but this can only give information on the correlation between the two measures, and is not informative to the question of whether or not the two measures can be considered equivalent. Significant correlation cannot be considered evidence for significant equivalence. 120

Adapting the revised quality assessment of diagnostic accuracy studies checklist for objective A(1)

Where appropriate, the quality of diagnostic accuracy studies was assessed using the revised quality assessment of diagnostic accuracy studies (QUADAS-2). 121 For reasons explained below, QUADAS-2 was adapted for objective A(1) (see Appendix 5).

The QUADAS-2 is a broad tool used to assess the quality of primary diagnostic accuracy studies. For this part of the review we were interested in analytic validity of the test only (i.e. its accuracy and reliability in measuring 5-FU plasma levels). Whether or not the test can accurately predict patients’ response to and side effects of treatment (its clinical validity) and be implemented to improve patient outcomes (its clinical utility) are considered in objectives B–D. We adapted the signalling questions in the QUADAS-2 tool for use with laboratory analytical studies. This was informed by the Analytic validity, Clinical validity, Clinical utility and Ethical guidance for assessing analytic validity for genetic tests. 122

In domain 1 (patient selection), one signalling question, ‘Was a case–control design avoided?’, was removed. In this measure of analytic validity the outcome of interest (5-FU plasma level) is continuous, and therefore by definition there were no cases or controls. The focus of concerns regarding applicability was adapted from relating entirely to patients to also including to plasma sample concentrations.

In domain 2 (index tests), the signalling question ‘Were the index test results interpreted without knowledge of the results of the reference standard?’ was removed because the index test is objective. The signalling question ‘If a threshold was used, was it pre-specified?‘ was removed as we were interested in agreement between two continuous measure without a threshold. An additional signalling question was added to account for the potential bias in under-reporting or not including failed tests: ‘Were the number of failed results and measurement repeats reported?’. Under applicability we added ‘Describe the preparation and storage of the sample before the index test was applied’ to check whether or not sample preparation was similar to potential NHS practice.

In domain 3 (reference standard), the signalling question ‘Were the reference standard results interpreted without knowledge of the results of the index test?’ was removed because the reference standard is objective.

In domain 4 (flow and timing), exclusions from the ‘2 × 2 table’ and ‘analysis’ were replaced with exclusions from the ‘Bland–Altman plot’ because there will be no thresholds used and therefore no 2 × 2 tables produced, and the outcome of most interest is the Bland–Altman plot (see Data extraction strategy). Additionally, ‘Did all patients receive a reference standard?’ was replaced with ‘Were both index test and reference standard conducted on all samples?’

Quality assessment strategy for objectives B and C

For objectives B and C, as a broad range of study designs were identified in the scoping searches, the use of a single checklist, in contrast to individual checklists for each study design, was considered appropriate. The Downs and Black checklist123 was therefore used to assess the quality of papers meeting the inclusion criteria (see Appendix 6). This 27-item checklist enabled an assessment of randomised and non-randomised studies and provides both an overall score for study quality and a profile of scores not only for the quality of reporting, internal validity (bias and confounding) and power, but also for external validity. However, as some questions were not appropriate for single-arm studies, the overall score was not considered useful or appropriate and was therefore not used. The results of the quality assessment provide an overall description of the quality of the included studies and provide a transparent method of recommendation for design of any future studies. Quality assessment was undertaken by one reviewer and checked by a second reviewer, any disagreements were resolved by a third reviewer through discussion.

Diagnostic accuracy studies (My5-FU vs. high-performance liquid chromatography/liquid chromatography-mass spectrometry) [objective A(1)]

The My5-FU assay delivers an estimate of plasma 5-FU concentration. For a study population this may potentially allow discrimination of study populations into categories: overdosed, optimally dosed and underdosed. Where results from a gold standard were available, a 2 × 2 table was constructed allowing diagnostic accuracy to be estimated using standard statistics (e.g. sensitivity, specificity, positive and negative likelihood ratios, positive and negative predictive values).

Diagnostic accuracy studies (My5-FU vs. HPLC/LC-MS) are considered to be those where patient samples are assayed for 5-FU concentration but patient outcomes may not be reported. Those studies that aimed to test the internal and/or external validity of the My5-FU assay were identified and their findings were summarised and appraised. Studies that do not report test failure rates were noted; where available, test failure rates were tabulated.

Patient-based studies (objectives B and C)

Analyses was stratified according to cancer type, 5-FU delivery mode and cancer stage (e.g. metastatic).

Study, treatment, population and outcome characteristics were summarised and compared qualitatively and, where possible, quantitatively in text, graphically and in evidence tables. Pooling studies results by meta-analysis was considered. Where meta-analysis was considered unsuitable for some or all of the data identified (e.g. due to the heterogeneity and/or small numbers of studies), we employed a narrative synthesis. This involved the use of text and tables to summarise data allowing the reader to consider any outcomes in the light of differences in study designs and potential sources of bias for each of the studies being reviewed. Studies were organised by research objective addressed. A commentary on the major methodological problems or biases that affected the studies was included, together with a description of how this may have affected the individual study results.

For objectives B and C we aimed to identify studies which compared BSA-based dose regimens of 5-FU with continuous infusion in which measures of plasma 5-FU are not undertaken to inform dose changes with dose regimens in which dose adjustment is informed by the My5-FU assay results applied to a stated dose adjustment algorithm. These studies would best report the following outcomes: incidence and severity of side effects of 5-FU; OS; and PFS, as stated in the inclusion criteria. We considered using a linked evidence approach124 in which studies report dose adjustment informed by plasma 5-FU measured by other methods (e.g. HPLC, LC-MS); this required a narrative linking of evidence of comparable performance of My5-FU with such assay methods.

In studies where My5-FU had been used but there was no comparator arm, or the comparator arm was a convenience sample (retrospective/historical population), outcomes were listed and appraised. Outcomes reported for non-randomised comparator arms (i.e. historical controls) were assessed for their representativeness in the light of information gained from systematic reviews (objective D). Relevant clinical outcomes from single-arm studies were considered for pooling should they be reported in sufficient detail and be considered relevant to the objectives.

Outcomes reported

Generally, the results confirm that higher levels of plasma 5-FU are related to improved outcome in terms of response, PFS and OS irrespective of the 5-FU regimen used. Furthermore, they appear to suggest that unfortunately the positive relationship between exposure to 5-FU and outcomes is stronger for AEs/toxicity than for response and survival. AEs were generally reported well; however, they were reported as risk (of experiencing at least one event) in 15 studies130,131,134,136,137,140,142,144–150,154 and as counts of events in seven studies. 133,138,139,143,151–153 Furthermore, studies that used different grading tools to grade severity of toxicity [WHO and National Cancer Institute Common Terminology Criteria for Adverse Events (NCI CTCAE) grouped grades of toxicity differently; some reported toxicity only for broad categories (i.e. haematological and digestive), others reported toxicities only as a total of all toxic events. Response rates varied hugely and median OSs from PK and BSA studies overlapped. Survival data in the form of Kaplan-Meier curves was only reported in three CRC studies130,134,138 and one H&N cancer study. 152

Quality concerns

Overall, the evidence from the single-arm studies is weak. Study conclusions were mainly based on small study populations. The majority of studies,130,131,133–135,137–141,143,147–150,152–154 were case series (18/24), which are generally of lower quality because selection bias cannot be assessed.

Evidence taken forward to the analyses and cost-effectiveness modelling

The emphasis here is in evidence that compares PFS, OS and toxic events for PK versus BSA treatments; these are crucial for an estimate of clinical effectiveness and for informing an economic model. Due to the heterogeneity among the studies, even within cancer types the studies do not lend themselves to pooling. Although 11 studies131,133,134,136,138,141,144,145,147,151,152 were identified that carried out dose adjustment, a comparison with the studies resembling the BSA arm is impossible due to the substantial differences in patient populations, treatment regimens and outcomes assessment (ways in which outcomes were graded, grouped and reported). Therefore, conclusions on the effectiveness of PK dosing cannot be inferred from the single-arm studies because of the lack of comparative evidence of PK versus BSA for any of the cancers. Survival data were reported for an obsolete H&N cancer treatment152 and for three CRC studies. 130,134,138 It was unclear whether the two studies by Gamelin et al. 130,138 used different populations, therefore only the later one was considered. Inferring Kaplan–Meier curves from single-reported medians would result in endless possibilities of different curves. Furthermore, it would be problematic to infer a curve from single-arm studies in the absence of complementary evidence from a comparative arm. Due to the clearly pronounced heterogeneity in reported medians in survival and the lack of CIs meant that no attempt was made at pooling values from single-arm studies. As a result, the usable evidence from this section comes from the two CRC studies which reported survival data in the form of Kaplan–Meier curves. 134,138 These supplemented the comparative CRC studies drawn from the following section on comparative studies and are considered further in Single-arm colorectal cancer studies taken forward in cost-effectiveness analysis. 134,138

Major features colorectal cancer studies

Three comparative CRC studies were identified. 118,134,155 These were published by Gamelin et al. ,118 Capitain et al. 119 and Kline et al. ;155 they were disparate with regard to design, population, treatments and reported outcomes. The major features of these studies are summarised in Table 11. Full data extraction forms are available from the authors on request.

Outcomes reported

Kaplan–Meier plots were reported variably by Kline et al. 155 for PFS; by Capitain et al. 119 for PFS and OS, but only for the PK arm; and by Gamelin et al. 118 for OS only. Response was reported by two of the three studies and AEs were reported as risks of experiencing at least one event.

Quality concerns

The only RCT118 did not report methods of randomisation and had perfectly balanced arms. In the other two studies,119,155 the absence of randomisation means that true comparability between groups is inevitably compromised. There is a further problem when patients are invited to self-select into PK or BSA dosing. See Appendix 11 for the Downs and Black123 quality assessment.

Evidence taken forward to the analyses and cost-effectiveness modelling

All three comparative CRC studies had useable information about PFS, OS and AEs/toxicity for the comparison of PK versus BSA treatment and are considered further in Comparative colorectal cancer studies taken forward in cost-effectiveness analysis.

Outcomes reported

Survival data for OS and PFS was not reported in either of the two studies. Information on response was provided and toxicity was reported as counts of toxic cycles.

Quality concerns

The only randomised evidence available for H&N cancer was hampered by mismatches between the description of methods undertaken and the reported results. Furthermore, since the patients with protocol violations were removed from the analysis and the induction therapy regimen used only two drugs, the generalisability to dose adjustment methods in current clinical practice remains questionable. See Appendix 11 for the Downs and Black123 quality assessment.

Evidence taken forward to the analyses and cost-effectiveness modelling

The studies by Santini et al. 132 and Fety et al. 156 date back to 1989 and 1998, respectively, and are the only comparative studies comparing BSA versus PK identified for H&N cancer. The two studies used regimens which are no longer in clinical use and did not provide estimates for OS and PFS. Fety et al. 156 provided some information on toxicity for the comparison of PK versus BSA dosing in H&N cancer patients and is therefore further considered in Comparative head and neck study taken forward in cost-effectiveness analysis. Further detail on the study by Santini et al. 132 is provided in Appendix 12.

Study design and quality

Capitain et al. 134 carried out a case series of 76 patients treated with 5-FU for advanced CRC and most had not received previous 5-FU treatment. The study included two regimens weekly or every 2 weeks of folinic acid and FUFOL. PK dose adjustment was based on plasma 5-FU measurements determined by HPLC and dose adjustment followed Gamelin et al. ’s130 dose algorithm. The median follow-up was 3.5 years. The study aimed to determine simple genetic factors that may aid the tailoring of 5-FU administration in first-line chemotherapy of advanced CRC.

The study was a case series where it was impossible to assess whether or not the study population was representative of the population from which the participants were recruited. Information on recruitment was minimal. There were weaknesses in the clarity and presentation of data. OS and PFS were reported including a Kaplan–Meier plot for OS, but without numbers-at-risk tables. Although the study reported AEs as risks they were not reported separately for the two different regimens included (see table 4 of the publication). 134 The study lacks information on plasma measurements, frequency of dose adjustment and outcomes of dose adjustment for the purpose of this review. See Appendix 11 for the Downs and Black123 quality assessment.

The study by Gamelin et al. 138 is a prospective case series involving 152 patients with mCRC from nine different centres. The median length of follow-up was 3 years. 5-FU therapy with individual dose adjustment was investigated in terms of efficacy, tolerance and survival in mCRC patients. The primary and secondary efficacy end points were survival and response rate respectively. PK dose adjustment was based on plasma 5-FU measurements determined by liquid chromatography and dose adjustment followed Gamelin et al. ’s130 dose algorithm.

The study was a case series where again it was impossible to assess whether or not the study population was representative of the population from which the participants were recruited. Patients lost to follow-up were not accounted for in the analysis which was not stated to follow an intention-to-treat (ITT) analysis. The AE rates were reported by cycles (counts); however, the number of total cycles is unknown for the 3-month period. Response rates, OS and PFS were reported extensively, including duration of response and Kaplan-Meier curves but no numbers-at-risk tables. See Appendix 11 for the Downs and Black123 quality assessment.

Population

The reported demographic characteristics of patients are summarised in Table 13.

Intervention

The two studies investigated dose adjustment of different 5-FU treatment regimens which are detailed in Table 14.

All patients outside the target plasma concentration of 2000–3000 µg/l138 or an AUC of 25 mg × hour/l134 received dose adjustment following a previously published algorithm130 (see Table 6). Patients with grade II toxicity received a dose reduction by 100 mg/m²,138 or by 10%. 134 In patients with grade III toxicity, treatment was interrupted and once toxicities were resolved, restarted with the dose reduced by 250 mg/m²,138 or by 25%. 134 Treatment was stopped for patients with grade IV toxicities.

Outcomes

Capitain et al. 134 reported that 9 out of the 76 patients were at high risk of 5-FU toxicity due to abnormally low 5-FU plasma clearance. Of these, three had known DPD polymorphisms. The objective response rate was 32.9%, with 6.6% of patients having CRs. Objective response rate was defined according to the Response Evaluation Criteria in Solid Tumours (RECIST) group. The median OS (Figure 8) and PFS were 20 months and 3.3 months respectively. AEs were reported as risks of having at least one toxic event. The most common side effects (all grades) were diarrhoea (22%), hand and foot syndrome (18%) and mucositis (7.5%). A total of 10.5% of all toxicities were grade III and IV, toxicities occurred in 10.5% of patients and nine patients were identified to be at high risk of toxicity due to low clearance of 5-FU. Certain genotypes were linked to toxicity and shorter OS. The authors presented genetic factors which warrant investigation in future clinical trials to determine patients at risk of 5-FU toxicity or resistance before treatment commences.

FIGURE 8.

Reconstructed Kaplan–Meier plot for OS for Capitain et al. 134

In Gamelin et al. 138 the mean 5-FU dose after 3 months of treatment was 1803 ± 386 (950–3695) mg/m². After cycle 1 only 6 (4%) patients had 5-FU measures in the target range. Under- and overdosing occurred in 124 (82%) and 14 (9%) patients respectively. After dose adjustment the 5-FU target range was reached in 143 (94.1%) patients. OS for all patients was 19 months and PFS was 11 months (Figure 9). Overall response rate in patients with measurable disease was 66/117 (56.4%) of which 18 (15.4%) had CR (as according to RECIST). Duration of response from the start of treatment to the time of disease progression was 17 months for CR and 20 months for PR (range 1–36 months). The number of cycles required to reach optimal therapeutic levels was significantly associated with level of response (for CR + PR vs. SD + PD, p = 0.05; and for CR + PR + SD vs. PD, p = 0.029). The majority of toxic events (all grades) were diarrhoea (39%) and hand and foot syndrome (30%), which were reported as counts of events.

FIGURE 9.

Reconstructed Kaplan–Meier plot for OS and PFS for Gamelin et al. 138

Study design and quality

Gamelin et al. 118 was a Phase III multicentre RCT which randomised 104 patients to a PK-adjusted FUFOL regimen and 104 patients to a BSA-based dose regimen. There were five centres all located in France. PK adjustment was achieved using HPLC-determined plasma 5-FU estimates coupled with a published dose adjustment algorithm. Median follow-up was 3 years. The pre-specified primary outcome was response rate according to RECIST criteria. Methods of randomisation and allocation concealment were not reported. There was no mention of stratification (e.g. according to performance status) or of the use of minimisation methods, yet arms were perfectly balanced for number of patients. Blinding to treatment was not possible; assessment of response rates was assessed by a panel of two independent radiologists and may have been blinded, but this was not specified. There was some mismatch between the description of methods undertaken and the reported results.

Capitain et al. 119 was a retrospective ‘proof-of-concept study’ comparing 118 patients who received a PK-directed dose-adjusted FOLFOX6 regimen with 39 patients who received a conventional BSA-directed FOLFOX6 regimen. PK adjustment was by HPLC-determined plasma 5-FU estimates coupled with a commercial adjustment protocol. The intervention patients came from eight centres and the comparator patients from two further and different centres. Median follow-up for PK patients was 1426 days (3.9 years; range 2.2–8.3 years) but was unreported for the BSA arm. The most important weakness in this study was that although the sampling frame for selecting the study populations was described, the proportion of eligible patients that was finally included was not reported; thus, the selection method was unclear. To what extent the study was prospective was unclear.

Kline et al. 155 was a small retrospective single-centre analysis of stage II/III and of stage IV CRC patients who self-selected for PK-adjusted (n = 38) or BSA-based (n = 46) FOLFOX6 or FOFIRI regimens. The numbers of patients receiving FOLFOX6 or FOFIRI were not reported. Median follow-up was 14 months (BSA) and 17 months (PK) for stage IV patients, and 23 months (BSA) and 16 months (PK) for stage II/III patients. Lack of randomisation was the major limitation. The allocation of treatments by patient self-selection increases the likelihood of allocation bias.

Populations

The reported demographic characteristics of patients are summarised in Table 15.

Intervention/comparator

All three studies118,119,155 compared BSA-based dosing with PK dose adjustment based on steady state plasma 5-FU levels. Different treatment regimens were used (FUFOL, FOLFOX6, FOLFIRI); these are summarised in Table 16, which indicates the dose of 5-FU at the first cycle.

Kline et al. 155 measured steady state plasma 5-FU with My5-FU, the other studies used HPLC. At the start of BSA-based therapy the same dose is applied for all patients, dose change only occurs when necessitated by toxicity and dose increases are not undertaken. Only Gamelin et al. 118 provided details of the PK dose adjustment algorithm; this is summarised in Table 17.

In the PK arm, Capitain et al. 119 adjusted dose according to an unreported commercial protocol; it is unclear if adjustments were guided by factors additional to plasma 5-FU. Kline et al. 155 used an algorithm supplied by the My5-FU manufacturer but details were not reported. In the PK arms both dose increases and decreases were implemented.

Outcomes: overall survival

Gamelin et al. 118 reported Kaplan–Meier analysis of OS; the plots appeared unusual implying that information for time of death was aggregated at spaced time intervals. The reconstructed Kaplan–Meier estimate using the method of Guyot et al. 125 is shown in Figure 10 and closely overlaps the published figure (available on request from authors).

FIGURE 10.

Reconstructed Kaplan–Meier plots for OS (Gamelin et al. 118).

Reported median survival was 16 and 22 months for BSA and PK arms, respectively, and the log-rank test for equivalence was p = 0.08. The median estimates appear sensitive to the long horizontal steps that occur at around 0.5 survival; it appears the 22 month estimate reported may be an overestimate.

Weibull, log-normal and log-logistic distributions provided satisfactory models for both arms (see Appendix 13). Cox proportional hazards regression for the reconstructed data provided a HR of 0.82618 (95% CI 0.6198087 to 1.101265). The log-rank test for equivalence provided a p-value of 0.18. A Weibull model assuming proportional hazards generated a HR of 0.829255.

Capitain et al. 119 reported 28 and 22 months median OS for PK and BSA arms respectively. No CIs were reported. A Kaplan–Meier plot for only the PK arm was published. The reconstructed plot (using the Guyot et al. 125 method) is shown in Figure 11 and closely overlaps the published figure (available on request from authors).

FIGURE 11.

Reconstructed Kaplan–Meier plot for OS of the PK group (Capitain et al. 119).

To investigate the reported OS difference between arms Weibull distributions were used with assumed proportional hazards (i.e. the PK shape parameter was retained for both arms); the median for the BSA arm was used to estimate the BSA scale parameter.

The resulting curves are shown in Figure 12(a). The HR for these was 0.586 (shape parameter 1.6691, scale parameters 0.002333 and 0.0039833 for PK and BSA respectively). The log-normal distribution also provided a good fit to the reconstructed PK OS data and was used to derive a log-normal estimate of BSA survival that satisfied the reported median of 22 months. Since the mu (µ) parameter defines the median, the PK sigma (σ) parameter was retained for the BSA arm and a new mu parameter found for BSA [here survival is given by: 1 – ɸ(ln(t) – µ)/σ; and median = exp(µ)]. The resulting curves are shown in Figure 12(b); this also shows the resulting HR which with this distribution is non-proportional between treatments. Table 18 summarises the model parameters and median and mean survival for the models.

FIGURE 12.

Overall survival modelled on Weibull and log-normal distributions (data from Capitain et al. 119).

The study by Kline et al. 155 did not report OS.

Outcomes: progression-free survival

Although the methods section in Gamelin et al. ’s study118 stated that PFS was analysed, no medians or Kaplan–Meier plots were presented. Requests to authors for IPD for PFS failed to illicit any data.

Values were reported for mean time spent in response categories (CR, PR, SD; as defined according to RECIST) and are summarised in Table 19. They infer a benefit from PK for PFS (i.e. the mean time for each response category prior to progression was greater for the PK arm than the BSA arm and involved a larger proportion of patients).

If it assumed (scenario A) that each patient proceeds from each response category directly to the progressed state then the mean time to progression is 6.0 and 7.5 months in the BSA and PK arms respectively (we assume means, rather than medians were reported because the durations have reasonably normal distributions). An alternative (scenario B) may assume that if patients in each category proceed to the next category (e.g. CR to PR and then to SD) before reaching the progressive state and the mean durations are as reported, then mean time to progression is 8.27 and 12.48 months in the BSA and PK arms respectively.

Given a mean time to progression under the assumption of normal distribution of duration times for each response type, it is possible to calculate Weibull parameters for a parametric model of time to progression using the following relationship:

where λ and γ are scale and shape parameters, respectively, and Γ represents a gamma distribution. Using the above, and given the mean, there are many solutions for shape and scale parameters unless one is fixed. In order to obtain a single solution, the further assumption is required that the shape parameter for PFS is the same as that for OS (these values were 1.827686 for the PK arm and 1.54066 for the BSA arm). With these assumptions the Weibull parameters for PFS in the PK and BSA arms under scenarios A and B are as shown in Table 20.

Similarly for a log-normal distribution description of PFS the relationship between mean and median may be used: mean = median × exp([σ²]/2). This has several solutions unless either median or σ is fixed. In order to obtain a single solution the further assumption was made that the σ parameter was the same as that for OS for the corresponding study arm (these values were 0.648108 for the PK arm and 0.6944542 for the BSA arm). Median relates to µ parameter according to median = exp[(σ × (normsinv(0.5))) + (µ)]. Thus, the µ parameter for PFS can also be obtained. With these assumptions the log-normal parameters for PFS in the PK and BSA arms under scenarios A and B are as shown in Table 21. Figure 13 summarises the resulting models of PFS.

FIGURE 13.

Progression-free survival for PK and BSA arms on Weibull (upper) and log-normal (lower) derived parameters under scenario A (figures a, c, e and g) and scenario B (figures b, d, f and h) (Gamelin et al. 118).

Capitain et al. 119 reported median PFS of 16 months and 10 months in PK and BSA arms respectively. CIs were not reported. A Kaplan–Meier plot was provided for only the PK arm. The reconstructed PK Kaplan–Meier (using the Guyot et al. 125 method) is shown in Figure 14 together with that for OS for the PK arm. Both plots closely overlap the published figures (available from authors on request).

FIGURE 14.

Reconstructed Kaplan–Meier plots for OS and PFS of the PK group (Capitain et al. 119).

Median PFS from the reconstructed PK arm plot was 16.0 (95% CI 12.0 to 20.0) months close to the reported value. Weibull, log-normal and log-logistic distributions provided moderately well-fitting models to reconstructed data (see Appendix 14). To obtain an estimate of PFS for the BSA arm the reported median of 10 months (based on 39 BSA arm patients) was used with the same procedure as described above for OS. For the Weibull model, proportional hazard was assumed (HR 0.4817). Table 22 summarises the model parameters for these models.

It is noticeable that the mean predicted with the log-normal and log-logistic models exceeds the means for OS shown in Table 22. This is probably due to the influence of the flatter part of the PFS Kaplan–Meier plot beyond 15 months. Figure 15 illustrates the modelled PFS. As log-normal and log-logistic models generate PFS means greater than OS means for the PK arm, they appear to be less appropriate than the Weibull. The availability of medians only for the BSA arm limits the reliability of the analysis.

FIGURE 15.

Progression-free survival modelled on Weibull and log-normal distributions (Capitain et al. 119).

Kline et al. 155 presented Kaplan–Meier analyses of PFS for stage II/III and stage IV patients. Reconstructed plots with the method of Guyot et al. 125 were discrepant from the published graphs; therefore censoring (tick) and event data were extracted from the plots and used to generate the illustrative graph shown in Figure 16.

FIGURE 16.

Illustrative graph of PFS. (a) stage II/III (NPK = 16, NBSA = 19); and (b) stage IV (NPK = 19, NBSA = 30) (Klein et al. 155). NBSA, number of BSA regimen patients; NPK, number of PK regimen patients.

For stage IV the test for equivalence of BSA versus PK (log-rank test) yielded p = 0.16. Median PFS was reported to be 14 months and 10 months for PK and BSA groups, respectively, with median follow-up of 22 months and 20 months in BSA and PK groups respectively.

For stage II/III patients the reported log-rank test for equivalence yielded p = 0.0429, suggesting delayed progression for the PK group.

Outcomes: adverse events

Gamelin et al. 118 reported the percentage of patients that experienced six categories of AE categorised according to four WHO grades of severity. Results for diarrhoea, mucositis, hand and foot syndrome, leucopenia, cardiac toxicity and conjunctivitis at 3 months and at end of treatment showed little difference; it is therefore assumed the data represents the risk of a patient experiencing the event at least once. Patient risk of diarrhoea, of hand and foot syndrome and of conjunctivitis was higher than for the other AEs; the results in each arm at end of treatment are summarised in Table 23.

The relative risk (PK vs. BSA) of AEs is summarised in Figure 17. The number of events was sufficiently small for cardiac toxicity and mucositis that any differences between treatments could be attributed to chance. For diarrhoea, and for leucopenia to a lesser extent, the PK regimen appeared to benefit patients. For hand and foot syndrome and for conjunctivitis risk was somewhat greater for patients receiving the PK regimen.

FIGURE 17.

Relative risk of AEs: PK vs. BSA (Gamelin et al. 118).

Capitain et al. 119 reported only four types of AE which fell within the National Cancer Institute’s (NCI’s) Common Terminology Criteria scale categories III or IV. 160 Surprisingly, hand and foot syndrome was not included. The results are summarised in Table 24.

The relative risk (PK vs. BSA) of grade III/IV AEs is summarised in Figure 18. It appears PK reduces risk of diarrhoea and mucositis and also neutropenia (although the latter not significantly).

FIGURE 18.

Relative risk of grade III/IV AEs (Capitain et al. 119). RR, relative risk.

The result for diarrhoea is similar to that reported in Gamelin et al. 118 where the relative risk for grade III/IV combined was 0.251 (95% CI 0.088 to 0.718). Unlike Capitain et al. ,119 Gamelin et al. 118 reported a similar risk of mucositis for each arm.

Kline et al. 155 evaluated toxicity graded according to the NCI CTCAE version 4.0 scale. Side effects which were considered grade III using the NCI CTCAE scale or were deemed sufficiently serious by the physician to warrant a dose reduction were ‘designated as adverse effects’.

Among stage IV patients, 37% in both BSA and PK groups experienced grade III toxicity (Table 25). Among stage II/III patients grade III toxicity was more common among BSA patients than PK patients (69% vs. 32%; p = 0.0437 by Fisher’s exact test).

Among all 19 CRC stage II/III patients receiving a BSA regimen, three experienced diarrhoea, vomiting or nausea, side effects associated more with 5-FU than the other components of chemotherapy; 8 of all the 16 CRC stage II/III patients who received PK regimens experienced these side effects (p = 0.0652; Fisher’s exact test). Among stage II/III patients, grade III toxicity was more common among BSA patients than PK patients (69% vs. 32%; p = 0.0437 by Fisher’s exact test). The incidence of different types of dose-limiting toxicities for both groups is listed in Table 25 according to stage of disease.

The number of treatment doses received before adverse side effects were observed was greater for PK-dose-adjusted patients than for BSA group patients; this applied for both stage II/III and stage IV patients. The results are summarised in Figure 19.

FIGURE 19.

Number of doses received before adverse side effects were observed. The horizontal line represents the median. Numbers of patients were 10, 6, 11 and 7 (Klein et al. 155).

Outcomes: response rates

Gamelin et al. 118 defined response rates as the primary outcome measure. Response rates are summarised in Table 26.

More patients in the PK arm than the BSA arm experienced CR and PR and fewer experienced progression. The relative risk according to response type is summarised in Figure 20.

FIGURE 20.

Relative risk of different types of response (Gamelin et al. 118). RR, relative risk.

In the study by Capitain et al. ,119 patients’ therapeutic response was assessed according to RECIST 1.1 criteria. Reported results are summarised in Table 27.

At 3 months response rates were superior for the PK group. The relative risk (PK vs. BSA) of different response categories is shown in Figure 21.

FIGURE 21.

Relative risk (PK vs. BSA) of response according response type (Capitain et al. 119). A relative risk > 1 indicates a favourable result for PK except for ‘progression’ where a relative risk < 1 indicates a favourable result for PK. RR, relative risk.

Response rates at 6 months for the PK arm, but not BSA arm, were also reported. These indicated that a CR was observed for 23 of 118 patients. At 3 and 6 months, 23% of both BSA and PK patients were classified as PD. This implies a difference of 3 months between arms when 77% remain un-progressed and this fits fairly well with the log-normal and Weibull models for PFS shown in Figure 15.

Kline et al. 155 did not report response rates.

Outcomes: dose adjustment and dose received

Gamelin et al. 118 reported that 94% of PK arm patients reached target range plasma 5-FU concentrations ( = 2500–3000 µg/l) in a mean of four treatment cycles. The dose received when in target range varied greatly between PK regimen patients (Figure 22). According to the algorithm (see Table 17), dose was adjusted in the PK arm if WHO grade II or III toxicities were experienced. Over the whole treatment period about 31% of PK arm patients had experienced grade II or III hand and foot syndrome (see Table 23). Consequently, it appears possible that an appreciable proportion of dose adjustments in the PK arm may not have been implemented in accordance with 5-FU AUC; this proportion was not reported.

FIGURE 22.

Weekly 5-FU dose received when in target range (PK group) and weekly dose received at 3 months (BSA-based regimen group); data read from graph and graph reconstructed (Gamelin et al. 118).

Most patients (≈ 85%) in the PK group had dose adjustment; the mean dose after 3 months of treatment was 1790 mg/m² (range 765–3300 mg/m²). Only 4 of 49 BSA-based regimen patients whose plasma 5-FU concentration was measured were within target range.

Most PK patients had their dose increased to levels above the starting dose of 1500 mg/m²; indicating that without dose adjustment these patients’ steady state plasma 5-FU concentration was judged to have been less than that desirable for full effect on cancer cells. The implication of this is that in the BSA arm (where dose was retained at 1500 mg/m² or was reduced because of toxicity), a substantial proportion of patients remained underdosed. According to the rationale of PK adjustment this might be expected to translate into reduced effectiveness in the BSA arm versus PK arm for outcomes such as PFS, OS and response rates. A smaller proportion of the PK group had dose reductions. Under the rational of PK adjustment this might be expected to translate into a more favourable toxicity profile for the PK arm than the BSA arm.

Gamelin et al. 118 reported that 5-FU plasma levels between 2.5 and 3 mg/l were correlated with grades I and II diarrhoea and grade I hand and foot syndrome. Grade III diarrhoea and hand and foot syndrome were reported to be associated with 5-FU plasma levels > 3 mg/l (3.5 mg/l and 3.9 mg/l, respectively; p = 0.02).

In the study by Capitain et al. 119 the PK group received a mean dose close to 2500 mg/m² (mean 94.32% ± 13.7% of theoretical dose), by 3 months most were receiving an adjusted dose, and dose increases and reductions of 10% and 20% from 2500 mg/m² were common. Thus, at 3 months, 56 of 118 patients received doses > 20% different from their starting dose. Dosage changes reported at 3 months for the PK group are summarised in Table 28. About 91% of PK patients required dose adjustment. About two-thirds of PK patients received dose increases and about 20% had their start dose reduced potentially translating into reduced toxicity compared with BSA patients.

Within 3 months the 5-FU dose was decreased by 15% ± 4% (range 10–25%) due to grade III toxic adverse effects in 4 out of 39 patients in the BSA-based dosage group.

Kline et al. 155 did not report the proportion of patients that received dose adjustment. A graph was presented illustrating the distribution of doses at each successive cycle. The median dose (horizontal line) remained the same at 2400 mg/m² across cycles irrespective of treatment regimen. Although BSA patients had dose reductions at increasing frequency with increasing cycles, in the PK group both dose increases and dose reductions were undertaken. Based on the published graph it appears that about 25–30% of 19 stage IV PK patients received dose increases by cycles 3 and 4. Some PK patients received dose reductions. Figure 23 summarises this for stage IV patients.

FIGURE 23.

Median dose and distribution of doses at successive cycles for stage IV patients. n = 30 and n = 19 for BSA and PK groups respectively (Klein et al. 155).

Figure 24 summarises the results for stage II/III patients. In the BSA group dose reductions and median dose decreased with increasing cycles. For the PK group the median dose remained unchanged and patients received dose increases and reductions in about equal proportions.

FIGURE 24.

Median dose and distribution of doses at successive cycles for stage II/III patients. n = 16 and n = 19 for BSA and PK groups respectively (Klein et al. 155).

Outcomes: performance status

Gamelin et al. 118 reported the influence of treatment on patients’ performance status in terms of the percentage of patients whose status remained unchanged, worsened or improved relative to baseline. The results are summarised in Table 29. A few patients improved performance status, but for most performance status remained unchanged or worsened.

Outcomes: plasma 5-fluorouracil clearance

Kline et al. 155 calculated 5-FU clearance by dividing the administered dose by the AUC measure of plasma 5-FU concentration (mg × hour/l). This estimate was interpreted as a measure of patients’ ability to metabolise 5-FU. The results indicated that clearance decreased as cycles of treatment accumulated, indicating a reducing ability to metabolise 5-FU. The results are summarised in Figure 25.

FIGURE 25.

Distribution of 5-FU clearance rates with increasing dose cycles (Kline et al. 155).

Outcomes: second-, third- and fourth-line treatments

Capitain et al. 119 was the only study that documented second-, third- and fourth-line treatments. The distribution of different types of post-first-line therapies was reported for the PK group only. These are summarised in Table 30. The proportion of patients that received second-, third- and fourth-line therapies gradually declined; the most common therapies used were FOLFIRI and targeted therapies.

Study design and quality

This multicentre RCT156 (involving three centres) assigned 61 patients to a PK-adjusted 5-FU regimen and 61 patients to a standard-based dose regimen. PK dose adjustment was based on 5-FU measurements determined by HPLC and dose adjustment followed a modified algorithm by Santini et al. 132 Owing to 5-FU-related toxicity, three patients (5.2%) in the BSA arm and three patients in the PK arm (6.1%) left the study before completing chemotherapy. The length of follow-up was unclear. The primary end point was the incidence of haematological toxicity and the secondary end point was the equivalence of disease response.

Randomisation was stratified by centre (three centres were involved). Methods of allocation concealment were not reported. Blinding to treatment was not possible; assessment of response rates was assessed by a panel of two independent radiologists and may have been blinded, but this was not specified. There was some mismatch between the description of methods undertaken and the reported results. There were weaknesses in the clarity and presentation of data. It has been previously noted by other authors85 that the dose adjustment method in this study may have been too complicated, as the 12 protocol violations in the treatment arm (12/61 patients enrolled) were all related to 5-FU dose adjustment miscalculations. Furthermore, as the patients with protocol violations were removed from the analysis and the induction therapy regimen used only two drugs, the generalisability to dose adjustment methods in current clinical practice remains questionable. See Appendix 11 for the Downs and Black123 quality assessment checklist.

Population

The reported demographic characteristics are summarised in Table 31. Patients had advanced H&N cancer and most had not received previous chemotherapy.

Intervention

Patients were assigned to receive three cycles of induction chemotherapy with cisplatin (100 mg/m² on day 1) and 5-FU (96-hour continuous infusion), either at standard dose (BSA arm: 4 g/m²) or at a dose adjusted according to the 5-FU AUC (PK arm).

Outcomes

Quality of life (QoL), OS and PFS were not reported. AEs were reported per cycle (counts). Among the 122 patients randomly assigned to one of the two treatment arms, 16 patients (13%) were found to be ‘unevaluable’ for response and toxicity (four patients in the BSA arm, and 12 patients in the PK arm). Grade II and IV neutropenia and thrombopenia were reduced in the PK arm when compared with the BSA arm (7.6% vs. 17.5%; p = 0.013). Mucosity (grades II and IV) was only observed in the BSA arm (5.1%). There was no significant difference in the objective tumour response rate between both arms (77.2% in the BSA arm vs. 81.67% in the PK arm).

5-Fuorouracil + folinic acid regimens: overall survival

Four CG1317 studies with usable Kaplan–Meier plots were identified in which patients with advanced mCRC received a BSA-based FUFOL regimen given by continuous infusion. 162,163,164,165 Reconstructed Kaplan–Meier plots are shown in Figure 27 together with the BSA arm for the Gamelin et al. 118 Note that infusion time used in Gamelin et al. 118 was different (shorter) than that of the other studies and that the plots are for single arms from different studies. The plots suggest that the control arm in Gamelin et al. 118 were not substantially different from the evidence available in the literature.

FIGURE 27.

Overall survival for Gamelin et al. :118 FUFOL BSA arm compared with studies from CG131. 7

Comparison of overall survival for body surface area arms with that for pharmacokinetic arms

Three studies provided Kaplan–Meier plots for OS for patients who received continuous infusion FUFOL regimens with PK adjustment of dosage; each was a publication from the same French investigative group. 118,134,138 Figure 28(a) summarises the reconstructed Kaplan–Meier plots from these studies. Note that although the treatment regimens were the same in the Gamelin et al. studies,118,138 in Capitain et al. ’s study134 they involved a much longer infusion time. The plots generate very similar median survivals. When plotted together with the BSA arms [see Figure 28(b)] there appears to be a small gain in OS deriving from PK adjustment. It should be emphasised that except for the contribution from Gamelin et al. 118 (for both PK and BSA arms) this is a non-randomised comparison of arms from separate studies.

FIGURE 28.

Reconstructed OS Kaplan–Meier plots for PK regimens. (a) BSA regimens; and (b) both PK and BSA regimens.

Baseline characteristics of the studies are summarised in Table 32.

If studies for each treatment arm are simply combined then Kaplan–Meier plots for each treatment appear as shown in Figure 29. It should be strongly cautioned that there are many caveats regarding the validity of this procedure, including the assumptions of similar treatments and similar populations. Furthermore, there is a lack of adjustment for potential patient- or study-level confounders. Parametric fits for these and for the individual studies are shown in Appendix 14.

FIGURE 29.

Kaplan–Meier plots resulting from combining OS data from studies: the plots and 95% CIs should be viewed with caution.

5-fluorouracil + folinic acid: progression-free survival

No PFS data was available for the BSA FUFOL regimen from the studies included in objectives B and C. Three CG1317 studies with usable Kaplan–Meier plots were identified162,163,165 (a further study by Giacchetti et al. 170 was excluded because the 5-day chrono-modulated continuous infusion employed was judged too dissimilar to the relevant regimen).

Figure 30(a) shows the reconstructed Kaplan–Meier plots for these studies. The plots are similar and generate medians that are very close to those in the published articles. Only one study138 from objectives B and C provided evidence about PFS of advanced CRC patients treated with a FUFOL PK regimen. When plotted with the CG1317 studies there is an apparent gain in PFS from the PK regimen. However, caution should be exercised as it should be born in mind that this evidence comes from single arms of independent studies. Table 33 summarises the baseline characteristics of the populations from these studies.

FIGURE 30.

Progression-free survival Kaplan–Meier plots for three BSA-based FUFOL studies (a) compared with that for one PK-based study (b).

5-Fuorouracil + folinic acid regimens: difference between overall survival and progression-free survival under body surface area and pharmacokinetic regimens

Figure 31 summarises the apparent difference between OS and PFS under BSA- and PK-based FUFOL regimens based on the evidence described above.

FIGURE 31.

Kaplan–Meier plots illustrating the difference between OS and PFS for FUFOL regimens based on BSA dosage and PK-adjusted dosage. (a) BSA-based regimens; and (b) PK-based regimens.

FOLFOX6 regimens: comparison of body surface area and pharmacokinetic arms overall survival

Four CG1317 studies with usable Kaplan–Meier plots were identified in which patients with advanced mCRC received a BSA-based FOLFOX6 regimen;164,166–168 in addition, our clinical advisors pointed to the existence of the UK COIN trial. 34 Reconstructed Kaplan–Meier plots for these five studies are shown in Figure 32. The two UK studies (Seymour et al. 164 and COIN as reported in Madi et al. 169) provide very similar OS that is somewhat less than the other three European studies. 166–168 Also shown is the reconstructed Kaplan–Meier plot for the PK arm of the comparative study of Capitain et al. 119 Unfortunately, Capitain et al. 119 only provided the median OS for the BSA arm (22 months). This corresponds closely to the median for the three non-UK BSA arms.

FIGURE 32.

Reconstructed Kaplan–Meier plots of OS comparing PK and BSA regimens.

The difference between the PK and BSA plots implies an OS advantage from PK adjustment of dosage. It should be emphasised, however, that the plots are for single arms from different studies and do not represent a randomised comparison; baseline characteristics for these studies are summarised in Table 34.

FOLFOX6 regimens: progression-free survival

Other than a reported median survival (10 months) without CIs no PFS data was available for the BSA FOLFOX6 regimen from the studies included in sections B and C. Two CG1317 studies with usable Kaplan–Meier plots were identified,167,168 in addition the COIN trial34 as reported by Madi et al. 169 was indicated to us by clinical experts. Figure 33 shows the reconstructed Kaplan–Meier estimates for these three trials and for the only available PFS data for FOLFOX6 with the PK-adjusted dosage (Capitain et al. 119). The median of 10 months for the BSA arm in Capitain et al. 119 is slightly greater than that for the three BSA plots (which were 9.3, 8.9 and 8.1 months in the reconstructed plots for Ducreux et al. ,167 COIN169 and Tournigand et al. 168 respectively).

FIGURE 33.

Progression-free survival Kaplan–Meier plots for three BSA-based FOLFOX6 studies compared with that for one PK-adjusted study.

The difference between the PK and BSA plots implies a PFS advantage from PK adjustment of dosage. It should be noted that the plots are for single arms from different studies and do not represent a randomised comparison and that the PK evidence comes from a single study. Demographic characteristics of the studies are summarised in Table 35.

FOLFOX6 regimens: difference between overall survival and progression-free survival under body surface are and pharmacokinetic regimens

Figure 34 summarises the apparent difference between OS and PFS under BSA- and PK-based FOLFOX6 regimens based on the evidence described above. The evidence for the PK-adjusted regimen comes from a single comparative study which did not provide Kaplan–Meier plots for the comparator BSA arm.

FIGURE 34.

Kaplan–Meier plots illustrating the difference between OS and PFS for FOLFOX6 regimens based on (a) BSA dosage and (b) PK-adjusted dosage.

Adverse events/toxicity

The authors of NICE CG1317 commented that ‘mixed treatment methods were not applied to toxicity data as there was insufficient evidence to inform the analysis’. For the de novo economic model in CG131 (comparing different treatment strategies), the risk associated with only three potential toxicities were estimated: febrile neutropenia, grade III/IV diarrhoea and grade III/IV hand and foot syndrome. They were selected on the basis of data availability and their likely impact on QoL. CG1317 did not report any toxicity data for a first-line FUFOL regimen. Data for first-line FOLFOX regimens, taken from the appropriate arms of various studies, was presented and is summarised in Table 36, where data from the Capitain et al. 119 comparative study is added for comparison.

The risk of diarrhoea in the Capitain et al. 119 BSA arm appeared similar to that estimated by authors of CG131. 7 It should be appreciated that CG1317 included FOLFOX4 as well as FOLFOX6 arms in their analysis. The difference between CG1317 and Capitain et al. 119 in the risk of neutropenia may be due to different definitions. The risk of hand and foot syndrome appears to be low, but was not reported by Capitain et al. 119

Cost search 1: cost-effectiveness of pharmacokinetic dosing and 5-fluorouracil

The search strategy developed for objectives A–C of the clinical effectiveness review (for methods, see Chapter 3, Searches for objectives A–C and Appendix 2) was also used to identify any published cost-effectiveness studies. This was considered appropriate because no study type filters were applied. Full copies of all studies deemed potentially relevant by clinical effectiveness reviewers were obtained and assessed by a health economist for inclusion.

Cost search 2: adverse events associated with chemotherapy (all cancers) – quality of life

A series of search strategies was devised to update and expand the literature review of Shabaruddin et al. 171 The search strategies were developed iteratively and are provided in Appendix 1. Searches were undertaken in MEDLINE and EMBASE. All records were screened for inclusion by an information specialist and checked by a health economist. Full copies of all studies deemed potentially relevant were obtained and assessed by a health economist for inclusion.

Cost search 3: adverse events associated with chemotherapy (all cancers) – resource use

A scoping search was undertaken to look for existing reviews. Some reviews of interest were identified, but no relevant overarching review was found. Therefore, a search strategy was developed based on the strategies used for cost search 2. The search strategies were developed iteratively and are provided in Appendix 1. Searches were undertaken in MEDLINE. All records were screened for inclusion by an information specialist and checked by a health economist. Full copies of all studies deemed potentially relevant were obtained and assessed by a health economist for inclusion.

Cost search 4: metastatic colorectal cancer/head and neck cancer – quality of life; and cost search 5: metastatic colorectal cancer/head and neck cancer – resource use

Searches 4 and 5 were developed iteratively, with reference to the search strategies of several published systematic reviews. 172–176 Searches for resource use were limited to English, humans and the UK perspective (by the addition of several currency and location terms). No date limits were applied. Searches were undertaken in MEDLINE (see Appendix 1). All records were screened for inclusion by an information specialist and checked by a health economist. Full copies of all studies deemed potentially relevant were obtained and assessed by a health economist for inclusion.

Model structure

Where data allowed, the preferred approach was to model the impact of PK dose adjustment using My5-FU assay compared with BSA dosing, using the clinical outcomes specified in the above and with a lifetime horizon. In the absence of such evidence, a linked evidence approach was adopted, linking My5-FU dose adjustment to other PK dose adjustment studies within the literature. It may have assumed equivalence between the My5-FU assay and other PK measures of plasma 5-FU (i.e. HPLC and LC-MS) if this appeared a reasonable assumption in the light of the clinical review. Model inputs may have utilised indirect treatment comparison results or NMA results to derive estimates of the clinical outcomes for the chemotherapy regimens relevant to current UK clinical practice. It was anticipated that this would be possible for mCRC, as outlined in more detail Chapter 4, Objective A(1): review of studies examining the accuracy of the My5-FU assay when tested against gold standard methods.

Although it was desirable to try to link evidence through to final survival outcomes, it should be recognised that due to data limitations this may be impossible for some cancers. Where this applied, the assessment still endeavoured to estimate the impact of My5-FU dose adjustment on test costs, treatment costs, side effect costs and the QoL impacts of side effects. This truncated analysis would be augmented by threshold analyses that estimate what, if any, additional impacts My5-FU would be required to have on PFS and/or OS for it to be cost-effective at conventional NICE willingness-to-pay (WTP) thresholds. The estimates of the additional survival were reported in terms of the absolute additional time required, with this being compared with estimates of the relevant current mean survival. It was also reported in the same metric as that used for the estimate of the impact of PK dose adjustment on OS in the mCRC modelling, in order to facilitate a comparison across clinical areas (e.g. as a relative risk or as a HR).

Necessary choices and definitions regarding the structure of the model depended on the findings from the literature review and consultation with clinical experts.

Issues relevant to analyses

During scoping, no end-to-end studies of the My5-FU assay were identified. Evidence was found relating to the validation of the My5-FU assay with LC-MS; the impact of 5-FU plasma levels on toxicity; PK variability of 5-FU when BSA dosing is used; and the impact of PK dose adjustment of 5-FU on survival. Studies comparing PK dosing with BSA dosing in mCRC were found that reported average 5-FU weekly doses, AE rates, PFS and OS with varying degrees of completeness. 118,119 These, together with other papers that may be identified during the literature searches, may provide sufficient information to enable estimation of the various clinical outcomes for mCRC. The papers’ authors will also be approached for the information about the outcomes that were ambiguously, partially or not reported for one or both arms.

Where clinical outcome estimates can be arrived at for My5-FU informed dose adjustment and BSA dosing, these will be the preferred basis of the modelling. The main model structure will be developed to favour these elements over those that may be drawn from a linked evidence approach or from expert opinion. This does not preclude more speculative model structures also being developed.

One-way sensitivity analyses will be performed for all key parameters and for parameters in the models which are based on expert opinion or lie within any more speculative linked evidence modelling. The appropriate model structure may also be subject to some uncertainty. Probabilistic modelling will be performed using parameter distributions instead of fixed values. It may be necessary to perform a number of probabilistic modelling exercises, given the uncertainty around parameter estimates that are based on expert opinion and the uncertainty around the most appropriate model structure.

Decision uncertainty regarding mutually exclusive alternatives will be reflected using cost-effectiveness planes and cost-effectiveness acceptability curves (CEACs) or frontiers.

Longer-term costs and consequences will be discounted using the UK discount rates of 3.5% for both costs and effects.

Health outcomes

Utility values, based on literature or other sources, will be incorporated in the economic model. QALYs will be calculated from the economic modelling.

Costs

Data for the cost analyses will be drawn from routine NHS sources [e.g. NHS reference costs, Personal Social Services Research Unit (PSSRU), British National Formulary (BNF)], discussions with individual hospitals and with the manufacturer.

Costs for consideration will include:

• cost of My5-FU testing

• cost of delivering 5-FU by infusion

• cost of side effects and 5-FU toxicity and their associated treatment or hospitalisation costs

• additional costs associated with changes to continuous infusion protocols.

Other costs for consideration may include:

• cost of second-line therapies

• palliative care and end-of-life costs.

Cost and resource use

Resource use will be estimated in line with the DAP programme manual:

• The perspective will be that of the NHS and PSS.

• The cost of the My5-FU assay will be requested from the manufacturer on the basis of this being nationally and publicly available, with additional confirmation of this sought from a UK laboratory currently using My5-FU.

• The base case will use list prices for the chemotherapy regimens, but as in the modelling for CG1317 the impact of discounted prices available to the NHS may also be explored.

• The above two bullets may be augmented with advice from the UK NHS centre currently using the My5-FU assay and other bodies such as the NHS Purchasing and Supply Agency.

• The effect of My5-FU on resource use in terms of physical units will be presented separately and then coupled with unit costs.

Literature review: economics

The literature review first reviews the only cost-effectiveness study of My5-FU for PK dose adjustment. A number of further literature reviews are undertaken and presented in full in the following appendices:

• Appendix 15 , Previous National Institute for Health and Care Excellence metastatic colorectal cancer assessments and Department of Health report

• Appendix 16 , Quality-of-life papers and metastatic colorectal cancer

• Appendix 17 , Metastatic colorectal cancer UK resource use literature review

• Appendix 18 , Previous National Institute for Health and Care Excellence assessments in head and neck cancer

• Appendix 19 , Head and neck cancer: other quality of life literature review

• Appendix 20 , Head and neck cancer: other UK resource use literature review

• Appendix 21 , Adverse events and resource use

• Appendix 22 , Literature review of quality of life and adverse events.

These are briefly summarised in what follows. Their contributions to the assumptions and parameter values for this assessment are highlighted but not reviewed in detail within the main body of the report.

Cost-effectiveness studies of pharmacokinetic 5-fluorouracil dosing in metastatic colorectal cancer

Becker et al. 178 (funded by Saladax) in an International Society for Pharmacoeconomics and Outcomes Research poster presentation summarising the results of a cost–utility modelling exercise that compared My5-FU dose adjustment with BSA dosing among patients with mCRC. This adopted a lifetime horizon, and discounted costs and benefits at 3.0%. A range of chemotherapy regimens were analysed, with My5-FU dose adjustment being compared with BSA dosing for:

• FUFOL

• FOLFOX4

• FOLFOX6

• FOLFIRI

• FOLFOX6 + bevacizumab

• FOLFIRI + bevacizumab.

The detail of the modelling as presented below is drawn from personal communication and an electronic copy of the model (Russell Becker, Russell Becker Consulting, Chicago, IL, USA, 2013, personal communication).

[Academic-in-confidence (AiC) information has been removed.]

(AiC information has been removed.)

(AiC information has been removed.)

(AiC information has been removed.)

Chemotherapy administration costs in previous National Institute for Health and Care Excellence metastatic colorectal cancer assessments

For the chemotherapy administration costs the majority of assessments adopted a disaggregate approach, separately costing individual elements such as line insertion, pharmacy preparation, administration and line flushing. The most comprehensive source appears to be the costing of TA93,30 with these also being used for TA212. 183 CG1317 is a slight outlier in terms of only applying the NHS reference cost for the delivery of complex parenteral chemotherapy. The current assessment will adopt the approach of TA93. 30 It will also source some of the inputs, suitably uprated for inflation, such as the cost of pharmacy preparation from TA9330 and TA212. 183

Duration of treatment in previous National Institute for Health and Care Excellence metastatic colorectal cancer assessments

Assessments have varied between assuming only 12 weeks of treatment or treatment until progression or unacceptable toxicity. Most assumed treatment would be as per the relevant trials, which was in effect until progression or unacceptable toxicity.

However, this is complicated by UK clinical practice being now in part informed by the relatively recently reported COIN trial, as reported in Adams et al. 34 and Madi et al. 169 This assessed the impact of intermittent treatment compared with continuous treatment, with patients receiving oxaliplatin plus 5-FU and capecitabine plus 5-FU with some crossing over between the two treatments. Intermittent treatment during COIN consisted of an assessment at 12 weeks. Those with PD moved onto off-protocol treatment, but those with SD or responding disease entered a chemotherapy-free treatment period with ongoing monitoring. If PD was found during this ongoing monitoring, patients recommenced their original chemotherapy. This intermittent treatment continued until PD was observed while on therapy or the patient chose to stop.

Expert opinion suggests that a treatment break of between 6 and 12 cycles of FOLFOX would be usual UK practice, with the responses suggesting that 12 cycles would be the more common goal. Given 12 cycles, it is also suggested that very few patients would enter a second course of treatment: at most 20% and more probably < 10%.

In the light of this, the current assessment will assume for the base case that the goal of treatment is to achieve 12 cycles of treatment. Progression before this point will limit the number of patients that actually achieve this goal.

There remains uncertainty about the actual proportion of PFS that UK mCRC patients spend receiving first-line chemotherapy. The National Cancer Intelligence Network (NCIN) Systemic Anti-Cancer Therapy Dataset184 should in time be able provide information about the chemotherapy regimens being used by tumour type and stage. The NCIN Chemotherapy Intelligence Unit was approached by the External Assessment Group (EAG) about obtaining information about the dosing and durations of first-line therapy for mCRC. This was with a view to then obtaining information along the lines of the following five bullets, for example mCRC patients receiving FOLFOX:

• the distribution between 5-FU doses during patients’ first cycles

• the split between those receiving only one continuous course and those receiving sequenced courses with treatment holidays in between

• the average course duration and standard error for those receiving only one continuous course

• the average course duration and standard error for those receiving sequenced courses with treatment holidays; and

• the average treatment holiday duration and standard error for those receiving sequenced courses with treatment holidays.

However, at present the NCIN Systemic Anti-Cancer Therapy Dataset184 is still in its infancy. Although some publications are available, the EAG was informed that the level of granularity and the completeness of the data are not sufficient to enable reliable information about the dosing and durations of first-line therapy for mCRC to be supplied. The Chemotherapy Intelligence Unit anticipates that the data will become sufficiently robust for analyses such as this in 2014/15.

Data from the COIN trial would have enabled the proportion of PFS spent on treatment to have been estimated. Unfortunately, despite the Trial Steering Committee agreeing to its release for this report, it was not possible to arrange an agreement with the University College London for this data to be released.

Modelling of progression-free survival and overall survival in previous National Institute for Health and Care Excellence metastatic colorectal cancer assessments

Buyse et al. 185 was quoted as concluding that tumour response is a weak predictor of overall response, summarising the results of the main systematic review of this. Pooling patient-level data from 3791 CRC patients enrolled in 25 RCTs suggested that only 38% (95% CI 9% to 69%) of the variation in OS was explained by variations in response rates. As a consequence, in common with the previous NICE assessments in mCRC the economics will restrict itself to modelling survival and PFS from parameterised curves reported in the clinical effectiveness section.

For the most part, parametric curves have been fitted to Kaplan–Meier curves. While not employing the method of Guyot et al. ,125 the Weibull was the most generally used, though CG1317 adopted the exponential. These curves have quite often been used to calculate the mean survival mathematically, rather than model it. Although compact, this method does mean that discounting was not applied. This was justified by the relatively short mean survival time, implying that discounting would have little effect on net quantities and the resulting cost-effectiveness estimates.

However, this is not obviously the case. For instance, the Guyot et al. 125 estimated OS Weibull for PK dosing as drawn from Capitain et al. 119 suggests a mean OS of 33.73 months. Applying the same shape parameter to the Capitain et al. 119 median OS of 22 months for BSA dosing suggests a mean OS of 24.48 months. A net gain from PK dosing of 9.25 months. But discounting at an annual rate of 3.5% reduces this net gain to 8.26 months: a reduction due to discounting of a little over 10%.

In the light of this, the current assessment will model the parameterised curves outlined within the clinical effectiveness section, with a 2-week cycle to reflect the duration of a cycle of FOLFOX. Continuous discounting, in the sense of each model cycle being associated with a unique discount factor to derive its present values, will be applied at an annual rate of 3.5% for both costs and benefits.

Quality-of-life values in previous National Institute for Health and Care Excellence metastatic colorectal cancer assessments

Previous NICE assessments in mCRC have typically applied two main QoL values: one for PFS and one for survival post progression. The literature underlying these estimates coupled with a systematic review of the literature in order to update these estimates is presented in Appendix 16. This can be briefly summarised as follows:

• TA61181 performed a cost-minimisation analysis so there were no QoL estimates.

• TA9330 drew a mean value of 0.76 from European Quality of Life-5 Dimensions (EQ-5D) data of the FOCUS trial (fluorouracil, oxaliplatin, irinotecan use and sequencing). 186

• TA118179 drew a value for PFS of 0.80 from the Ramsey et al. ’s187 Health Utilities Index (HUI) study among 173 US CRC survivors. A multiplier of 0.75 for survival with progression (SWP) was informed by the standard gamble (SG) survey among 30 UK oncology nurses of the Petrou and Campbell study. 188

• TA176182 drew a value of 0.79 for first-line treatment from EQ-5D data collected during the pivotal trial. Third-line therapy was assigned a value of 0.68 as drawn from Jonker et al. ,189 whereas second-line therapy was assigned a value at the mid-point of 0.73. SWP was assigned 75% of the value for first-line PFS.

• TA212183 largely relied on the value of TA176. 182

• CG1317 drew values of 0.51 for SD and 0.21 for PD from the Best et al. 190 time trade-off (TTO) study among 49 members of the US general public.

• TA242180 used values of between 0.75 and 0.81 for PFS and between 0.69 and 0.79 for SWP. These were supplied within the manufacturer submission, these being based on a reanalysis of the data underlying the Mittman et al. 191 HUI study of a Canadian publicly funded trial of adding cetuximab to the treatment of 575 mCRC patients.

The values used for CG1317 are outliers compared with the values used in other NICE mCRC assessments. Note that Best et al. 190 also surveyed 49 colorectal patients who reported values of 0.46 and 0.38 for the health states of metastatic SD and metastatic PD.

The systematic literature review only identified an additional four references that had not been considered at some point within the previous NICE mCRC assessments. Färkkilä et al. 192 is probably the most interesting, having surveyed Finnish mCRC patients using the EQ-5D value using the UK social tariff. Those with advanced disease were divided into patients with mCRC who were still receiving oncological care (n = 110) and those only receiving palliative care (n = 41). The mean QoL values were 0.820 [standard error of measurement (SEM) 0.019] and 0.643 (SEM 0.049) respectively.

Shiroiwa et al. 193 undertook a TTO study among members of the Japanese general public for mCRC health states that resulted in estimates more in line with those of Best et al. :190 0.59 for treatment with XELOX and no AEs and 0.53 for treatment with FOLFOX and no AEs.

Wang et al. 194 analysed EQ-5D data from an open-label trial of panitumumab being added to best supportive care for chemotherapy-refractive mCRC patients. The valuation of EQ-5D data is not made clear, but QoL values for those without AEs were 0.768 in the panitumumab arm and 0.663 in the best supportive care arm.

There is also an argument that the last few months of survival may be at a somewhat reduced QoL. Odom et al. 195 analysed the EQ-5D values from a trial of panitumumab being added to best supportive care for chemotherapy refractive mCRC patients. Unfortunately this used a US valuation for the EQ-5D, but it suggested that there was a reasonably linear decline in QoL over time among patients at this stage of treatment.

The above suggests that both the results of Färkkilä et al. 192 and those used in the modelling of TA176182 and TA212183 are reasonable estimates. They both suggest a similar QoL decrement of around 0.18 for the move from PFS to PD, though it has to be recognised that the Färkkilä et al. 192 0.643 relates to mCRC patients receiving only palliative care. In the light of this, the current assessment will apply QoL values of 0.820 for PFS and 0.643 for SWP as drawn from Färkkilä et al. 192 The values of TA176182 and TA212183 will also be applied as a reasonable sensitivity analysis, whereas the general public values of Best et al. 190 will be applied as a further sensitivity analysis.

There is a need to model second-line treatment for a proportion of patients, and PFS from this line of treatment. One approach is to apply the same QoL value for PFS from second-line therapy as for PFS arising from first-line treatment. However, it can be argued that PFS from second-line therapy would tend to be at a lower QoL than that arising from first-line therapy.

Only the modelling of CG1317 applied QoL decrements for AEs. These were drawn from the Lloyd et al. 196 SG exercise among 100 members of the UK general public, though related to metastatic breast cancer.

Due to AEs possibly being more central to the current assessment than to the previous NICE assessments in mCRC, a systematic review of the QoL impacts of treatment-related AEs has been undertaken. This is presented in Appendix 22 and further summarised below in the section on AEs and QoL.

The overall QALY impact of AEs also depends on the duration of AEs. It had been hoped that data from the COIN trial would have informed this for grade III and grade IV events. Although the COIN Trial Steering Committee was willing to release patient-level data for this report, it proved impossible to arrange a data release agreement with the University College London. In the absence of other data, for the current assessment this has been drawn from expert opinion.

Adverse event costs in previous National Institute for Health and Care Excellence metastatic colorectal cancer assessments

In general, where AEs have been costed a single aggregate AEs cost has been applied within NICE mCRC assessments, though this is at time differentiated by treatment arm. This is usually not sufficiently disaggregated to be applied to particular AEs as required for the current assessment.

Although not entirely transparent, based on NHS reference costs TA176182 estimated a cost per admission for a grade I/II AE of £1050 [£1216] and for a grade III/IV AE of £1170 [£1354].

TA212183 provides a more disaggregate costing of AEs. However, the details of this are not presented and despite the Evidence Review Group (ERG) asking for further information, none was apparently forthcoming.

CG1317 costs grade III/IV diarrhoea at £388 [£420] based on the NHS reference cost FZ45C short stay NHS reference cost.

In the light of this, for the current assessment the costs of individual grade I/II and grade III/IV AEs will be based on:

• the likelihood of hospitalisation

• NHS reference costs for those hospitalised

• medication costs for those not hospitalised.

The values for these are based on expert opinion as outlined in greater detail in the AEs sections below.

The modelling approach

The model has been constructed along the lines of a cohort distributed between health states over a 20-year time horizon. Given the survival curves of the clinical effectiveness section, this is in effect a lifetime horizon. A 2-week cycle has been employed to be in line with the FOLFOX cycle length, with half cycle correction in order to align survival estimates with those of the clinical effectiveness section. First-line treatment is assumed to be FOLFOX6. The distribution between health states is determined by the parameterised curves, with the first-line OS curve and the first-line PFS curve determining the post progression from first-line therapy curve. A constant proportion of those progressing from first-line therapy go on to receive second-line therapy which is assumed to be FOLFIRI. This is also associated with a PFS curve, as drawn from Tournigand et al. 168

The main health states of the model in terms of PFS, SWP and death, and the possible movements between them are outlined below (Figure 36).

FIGURE 36.

Metastatic CRC model structure.

In terms of the model structure, for the base case it is assumed that the default is for patients to move from PFS into SWP and then on to death. Moving directly from PFS to death is the exception. This only applies when there is an adding up constraint (i.e. the incident number of deaths implied by the OS curve exceeds the number of patients in the SWP health state).

The cost and QALY impacts of AEs associated with first-line treatment, differentiated by a My5-FU dose adjustment arm and BSA dose adjustment arm, have been included within the modelled, as outlined in greater detail below.

First-line therapy: survival estimates

Capitain et al. 119 studied PK dose adjustment using FOLFOX6, which is directly relevant to current UK clinical practice. However, there are problems with the control arm. Not only was it not randomised and somewhat smaller than the PK dose adjustment arm, only the median OS and median PFS are reported for it.

Gamelin et al. 118 provides the clearest analysis of the possible difference in OS using PK dosing. However its relevance may be hampered by the regimen being of questionable relevance to current UK practice: FUFOL rather than FOLFOX6. There is also no reliable reporting of PFS, though some inference may be made from the reported durations of response.

The parameterised Weibull OS curves differ noticeably between Gamelin et al. 118 and Capitain et al. 119 Figure 37 shows the first 5 years.

FIGURE 37.

Capitain et al. 119 and Gamelin et al. 118 PK Weibull OS curves for PK dosing.

In the light of this, two strands of analysis will be presented: one based on FOLFOX6 studies and one based on FUFOL studies. Within the strand based on FOLFOX6 studies a sensitivity analysis applying the HR for OS derived from Gamelin et al. 118 will be performed as a bridge. Note that within the strand based on FUFOL studies, owing to the regimens not being current standard UK practice, the drug costs will be based on FOLFOX6 for first line and FOLFIRI for second line.

The clinical effectiveness section concluded that an assumption of equivalence between My5-FU dose adjustment and PK dose adjustment by more traditional methods was justified reasonable assumption. As a consequence, the economic modelling will assume that the PK dosing curves of the literature are equally applicable to My5-FU dose adjustment. If this assumption is not valid, none of the results of the economic modelling hold.

FOLFOX studies analyses

The base case will apply the parameterised Weibull OS curve and overall PFS curve estimated from the PK dose adjustment arm of Capitain et al. 119 to the My5-FU arm. The Weibull OS curve and overall PFS curve for the BSA arm will be those estimated from the medians reported in Capitain et al. ,119 coupled with an assumption of them having the same shape parameter as the corresponding PK dose adjustment curve. This is reported in more detail in Tables 19 and 23.

Due to Capitain et al. 119 only reporting the median OS for the BSA arm, a scenario analysis will be presented that applies the proportionate hazard of OS derived from Gamelin et al. 118 of 0.829255, as reported in greater detail in Chapter 4, Outcomes: overall survival.

Due to the concerns around the size and derivation of the BSA control arm within Capitain et al. 119 and that only medians are reported, a range of scenario analyses that apply curves for the BSA arm derived from the wider literature will be presented. The studies underlying these estimates are reported in more detail in Chapter 4, FOLFOX6 regimens: comparison of body surface area and pharmacokinetic arms overall survival and Chapter 4, FOLFOX6 regimens: progression-free survival.

A scenario analysis will apply the Weibull OS curve estimated by combining the data from the five BSA studies [Seymour et al. ,164 Madi et al. 169 (COIN), Hochster et al. ,166 Ducreux et al. 167 and Tournigand et al. 168] and the PFS curve estimated by combining the data from the three BSA studies of FOLFOX [Madi et al. 169 (COIN), Ducreux et al. 167 and Tournigand et al. 168]. The curves of the base case will be retained for the My5-FU arm.

Two further scenario analyses will apply the Weibull OS curves and PFS curves estimated from (a) Ducreaux et al. 167 and (b) Tournigand et al. 168 The curves of the base case will be retained for the My5-FU arm.

The analysis of Chapter 4 suggests that the survival in the UK-based studies is somewhat worse than that of the other studies. This might imply that any analysis based on pooling data from single arms for the BSA arm that includes the UK studies may tend to bias the analysis in favour of My5-FU. A fairer comparison with the PK curves of Capitain et al. 119 might be to pool the data from single arms for the BSA arm, but excluding the UK studies. This will be undertaken as a final sensitivity analysis.

This leads to the following base-case analyses and scenario analyses (Table 44).

5-Fluorouracil + folinic acid studies analyses

Note that the following analyses still assume current UK practice in terms of the drug costs for first-line therapy being those of FOLFOX. The FUFOL study analyses apply the curves from the FUFOL studies, this being motivated, in part, by the only RCT of PK dosing being a FUFOL study.

Due to Gamelin et al. 118 not presenting the PFS curves, the base case will adopt the most conservative approach. It will apply the parameterised Weibull OS curves estimated from Gamelin et al. ,118 but it will assume equivalence between My5-FU and BSA dose adjustment for PFS, and apply the Weibull PFS curve estimated by combining the arms of the three main BSA studies. 167–169

Two scenario analyses will apply the Weibull PFS curves inferred from the durations of response reported in Gamelin et al. 118 as reported in more detail in Chapter 4, Outcomes: progression-free survival, while retaining the other curves of the base case.

A scenario analysis will apply the Weibull PFS curve estimated from Gamelin et al. 138 for the My5-FU arm, while retaining the other curves of the base case.

A scenario analysis will apply the Weibull OS curve estimated by pooling the results of Gamelin et al. 118,138 and Capitain et al. 134 in the My5-FU arm, while retaining the other curves of the base case.

A further scenario analysis will apply the Weibull OS curve estimated by pooling the results of Gamelin et al. 118,138 and Capitain et al. 134 in the My5-FU arm, the PFS curve estimated from Gamelin et al. 138 in the My5-FU arm, while retaining the curves of the base case for the BSA arm.

A final scenario analysis will apply the Weibull OS curve estimated by pooling the results of Gamelin et al. 118,138 and Capitain et al. 134 in the My5-FU arm, the PFS curve estimated from Gamelin et al. 138 in the My5-FU arm, the Weibull OS curve estimates from pooling five BSA studies,164,166–169 and the Weibull PFS curve estimates from pooling three BSA studies in the BSA arm. 167–169

This leads to the following base-case and scenario analyses (Table 45).

For the probabilistic modelling a Cholesky decomposition of the parameters variance-covariance matrix is applied.

Second-line therapy: survival estimates

The PFS among those receiving second-line chemotherapy is assumed to follow that of the second-line FOLFIRI of Tournigand et al. 168 (Figure 38). Analysis of this data suggests that both the log-normal and the log-logistic provide reasonable fits to this data (Table 46). However, there may be some concerns about extrapolating using these forms due to the relatively long tails they involve when extrapolating beyond the trial data. As a consequence, the base case will apply the Weibull parameterisation of the second-line FOLFIRI PFS curve, partly in order to increase consistency with the other curves that are being applied. The impact of this is likely to be slight, with all three parameterisations suggesting a mean PFS of around 0.34 years.

FIGURE 38.

Log-normal, log-logistic and Weibull for second-line FOLFIRI PFS (Tournigand et al. 168): over 50 2-week cycles.

Note that within this the duration, effect and cost of second-line therapy is treated as being independent of the duration, effect and cost of first-line therapy. This may not be accurate for even the deterministic analysis. If My5-FU increases PFS from first-line therapy compared with BSA dosing, this may also affect the duration, effect and cost of second-line therapy. The base-case assumption is that it does not. There is also no obvious data that would enable alternative assumptions about this to be parameterised.

Even with the base-case assumption that any increase in PFS from first-line therapy due to My5-FU dosing does not affect the duration, effect and cost of second-line therapy, a further problem may arise within the probabilistic modelling from modelling first-line effects and second-line effects independently. For some of the iterations of the probabilistic modelling the first-line PFS curve may be simulated as being quite close to the OS curve. This would squeeze the time spent in post first-line PFS, such that for some iterations there is insufficient time to accrue the simulated duration, effect and cost of second-line therapy. However, the mean second-line PFS of around 0.34 years is sufficiently short compared with the mean survival subsequent to first-line PFS for this not to be a concern.

Adverse events: rates

Adverse events rates are drawn from the key comparative papers. It should be borne in mind that these appear to report the proportion of patients experiencing events rather than the numbers of actual events. As a consequence, modelling underestimates the number of events to some degree, which could tend to bias the analysis.

As Capitain et al. 119 did not distinguish between grade III and grade IV AEs, where this distinction is required this balance will be drawn from the Gamelin et al. 118 data.

Progression-free survival and survival with progression quality of life

As previously reviewed in Quality-of-life values in previous National Institute for Health and Care Excellence metastatic colorectal cancer assessments, augmented with the results of the systematic review of Appendix 16, the results of Färkkilä et al. 192 and those used in the modelling of TA176182 and TA212183 are reasonable estimates. Both suggest a similar QoL decrement of around 0.18 for the move from PFS to PD, though it has to be recognised that the Färkkilä et al. 192 value of 0.643 relates to mCRC patients receiving only palliative care. In the light of this, the current assessment will apply QoL values of 0.820 for PFS and 0.643 for SWP as drawn from Färkkilä et al. 192 The values of 0.80 and 0.60 from TA176182 and TA212183 are also applied as a reasonable sensitivity analysis, whereas the general public values of 0.51 and 0.21 of Best et al. 190 are applied as a further sensitivity analysis.

Adverse events: quality of life

With the exception of CG131,7 the QoL impacts of AEs have not been separately modelled within previous NICE mCRC assessments. CG1317 included QALY decrements for AEs of 0.103 for grade III/IV diarrhoea, 0.150 for febrile neutropenia and 0.116 for hand and foot syndrome, as drawn from the Lloyd et al. 196 study of metastatic breast cancer.

In the light of this, a systematic literature review of the QoL impacts of chemotherapy-related AEs was undertaken, the results of this being reported in full in Appendix 22. This literature review updated and widened the literature review reported in Shabaruddin et al. 171 Based on just the values reported in this literature review, the following QoL decrements for grade III/IV AEs appear reasonable. Cardiac toxicity has not been further considered due to the only differences between the arms in Gamelin et al. 118 being for grade I/II cardiac AEs, these being asymptomatic (Table 47).

Due to the range of sources being drawn from for the above, clinical expert opinion was sought on the face validity of these estimates and their relative magnitudes. This can be summarised as viewing the QoL decrements for leucopenia, neutropenia and thrombocytopenia as too high. Apparently even for grade III/IV events, much of the leucopenia and neutropenia is asymptomatic. Similarly, thrombocytopenia was viewed as being unlikely to significantly affect QoL unless it led to bleeding. The probability of requiring platelet transfusion for thrombocytopenia was also viewed as being very small, suggesting that any QoL decrement associated with thrombocytopenia should perhaps be only applied to a small percentage of patients.

There was a suggestion that diarrhoea would have a larger QoL decrement than nausea/vomiting, whereas mucositis might have a smaller impact. Febrile neutropenia was agreed to have the largest QoL decrement.

The literature review identified the abstract by Boyd et al. 197 which reports the interim results from an analysis of the Medical Research Council (MRC)-funded Short Course Oncology Therapy (SCOT) trial201 of patients with fully resected stage III CRC or fully resected high-risk stage II disease. This presents the EQ-5D QoL decrements associated for a subset of both grade I/II and grade III/IV AEs. The interim analysis has been verbally presented in more detail by the author, with EQ-5D data being collected from 1292 patients at baseline, every cycle, 9, 12, 18, 24 months and annually thereafter. This data has been further analysed within a univariate analysis, with the following results (Kathleen Boyd, Glasgow University, 2013, personal communication) (Table 48).

The ‘not available’ values for the grade III/IV events were due to insufficient numbers of events: none for alopecia and photophobia, one for rash and watery eye, two for anaemia and mucositis clinical and six for thrombocytopenia.

The univariate analysis regressed the AEs against QoL individually. A multivariate analysis was also undertaken in which all AEs were simultaneously regressed against QoL. The multivariate analysis led to inconclusive results, with the individual AEs being highly correlated with one another. Dropping some AEs from the multivariate analysis still led to inconclusive results. In the effective absence of multivariate results, the authors recommend using the results of the univariate analysis.

The multicollinearity between the AEs of Boyd et al. 197 would appear to raise the possibility that within the univariate analyses the estimated QoL decrements for an AE are picking up not only the impact of the AE under consideration, but also some of the impacts of the AEs with which it is highly correlated. As a consequence the univariate values should perhaps be treated with some caution, with there being the possibility of double counting the QoL impacts of AEs.

The Boyd et al. 197 data still appears to be the best data in terms of robustness and alignment with NICE methods, but Lloyd et al. 196 also seems a credible source and unaffected by the problems of multicollinearity. In the light of this, the QoL impacts of AEs have been calculated on the following basis.

For grade III/IV AEs:

• diarrhoea taken directly from the SCOT trial201 univariate estimates

• nausea/vomiting taken as the mean of the SCOT trial201 univariate estimates

• hand and foot syndrome assumed to be greater than the diarrhoea decrement by the Best et al. 190 proportion

• mucositis taken from the SCOT trial201 univariate estimates

• neutropenia taken from the SCOT trial201 univariate estimates

• leucopenia assumed to be as per neutropenia

• thrombocytopenia taken from Swinburn et al. 200

However, in the light of expert opinion neutropenia, leucopenia and thrombocytopenia are assumed to have no QoL impact for the base case.

Table 49 includes QoL decrements for the grade I/II AEs. These are largely driven by the SCOT trial201 estimates, but as with the estimates for grade III/IV AEs they are also informed by the other estimates within the literature. For analyses based on the AE rates of Gamelin et al. 118 the QoL estimates for grade I/II AE rates were applied within sensitivity analyses. There remain concerns with the SCOT trial201 estimates for grade I/II AEs given the issues around multicollinearity.

Adverse event: durations

There is a paucity of data on the duration of AEs within the literature. The duration of AEs is as an important a driver of the QALY impact of AEs as the QoL decrements outlined above. As already noted, there is duration data for grade III and grade IV AEs within the COIN trial data set, but it was not possible to arrange an intellectual property agreement with the University College London.

Expert opinion suggests that the following may be reasonable (Table 50).