Individualising breast cancer treatment to improve survival and minimise complications in older women: a research programme including the PLACE RCT

Study 1

This study complemented the BCC project (protocol submitted to the NIHR with an original application reference of 2008NovPR35) focusing on patients’ perceptions of responsibility for the surgical decision. The BCC study was a prospective cohort study of 550 women aged ≥ 70 years consecutively recruited from newly diagnosed patients with operable (stage I–IIIa) breast cancer attending breast units in Greater Manchester over 21 months. The BCC study collected data on women’s preferences through an interview conducted at home. Study 1 complemented the BCC work by measuring surgeons’ perceptions of who made the treatment decision for the same index cases and related to the same consultations. Thus, we were able to collect a measure of agreed responsibility for treatment decisions. Regardless of whether an older woman received standard or non-standard treatment, we were able to establish whether this decision was a result of the patient’s or the surgeon’s choice. Data on surgeons’ perceptions of responsibility for the surgical decision for individual consultations had to be collected by brief, immediately post-consultation interviews, a resource-intensive method.

Study 2

As part of the research funded by the BCC, NIHR Fellowship and this programme, we planned to identify predictors of surgical risk using multivariate modelling of data from our cohort. For study 2, we planned to develop these predictors into a pre-treatment health assessment/screening tool to assess risk of adverse outcome (i.e. ‘fitness for surgery’). Once we had developed the tool, we planned a feasibility trial following Medical Research Council complex intervention framework and guidelines. However, our modelling revealed no significant clinically novel predictors of surgical risk and therefore we were not able to build a viable screening tool, and hence could not proceed to conduct the planned feasibility trial (see Modelling surgical risk). We thus consulted with programme board and proposed and received board approval for further follow-up of the cohort (IMPACT study) and several additional analyses of the data to investigate outcomes so as to investigate the impact of lack of treatments on older breast cancer patients in the UK. 4 In addition, we undertook an analysis looking at the relationship between congruence (patient getting the treatment decision-making style she preferred) and HRQoL at follow-up.

Introduction

For these studies we used our established cohort of patients aged ≥ 65 years and diagnosed with early-stage invasive breast cancer in 22 trusts in England from 1 July 2010 to 31 March 2013. The extent to which lack of surgery is explained by patient health and choice has been investigated using a range of pre-treatment health measures, tumour characteristics and demographics collected prospectively from patient interview and case note review. 1 However, follow-up of subsequent adjuvant treatment (radiotherapy and/or chemotherapy following surgery) and the impact of lack treatment on survival and long-term HRQoL were not within the remit, resources or timescale of this previous work.

Background

Older women in the UK experience the highest incidence and worst survival for breast cancer, and are less likely to have standard treatment. 1,3,4 The impact of lack of treatment on older patients’ survival needs to be investigated. There is good evidence that poor survival is a particular problem for older breast cancer patients in the UK. Møller et al. 4 found that the 5-year relative survival for women aged ≥ 80 years is 61% in the UK, compared with 74% in Norway and Sweden. They conclude that this ‘leads to important questions about the adequacy of care provided for the oldest patients’. However, Møller et al. did not investigate access to treatment on survival. Moreover, the proportion of patients with comorbidities/frailty and later-stage breast cancer increases with age, and both of these factors also affect survival, and so these variables should also be investigated/adjusted for.

Treatment for breast cancer is based on clinical trials that excluded older women. Moreover, recent trials specific to older patients have closed as a result of failure to recruit. 13 The deficit of evidence on the risks/benefits of treatment for the age group most affected by breast cancer remains. Given the increasing proportion of older people in our population,11,14,15 this presents a growing problem, and studies of older women’s response to therapy are required in order to provide patients, physicians and policy-makers with evidence on which to base decisions about treatment. Surgery is the mainstay of treatment for early breast cancer and yet rates reduce among those aged ≥ 75 years. Omission of surgery leads to lack of local control, particularly at 2 years post diagnosis. 16 Although the only previous trial investigating surgery versus no surgery for older breast cancer patients planned to investigate costs, it closed as a result of failure to recruit. 6

Aims

Workstream 1 set out to address the following research aims in the second set of studies:

• to investigate the extent to which primary surgery for older women with early breast cancer increases survival and HRQoL and is effective as measured by quality-adjusted life-years (QALYs)

• to investigate follow-up adjuvant treatment (radiotherapy and/or chemotherapy post surgery) for older breast cancer patients regarding:

  • the extent to which adjuvant treatment increases survival and HRQoL and is effective

  • the extent to which lack of adjuvant treatment can be explained by patient health and choice.

Methods

Table 1 and Figure 2 (see Appendix 1) summarise the outcome variables, explanatory variables and main methods for each of the above aims, along with a flow diagram specifying the logic and numbers of patients in the specific analyses performed. Data on our established cohort of 944 women aged ≥ 65 years consecutively diagnosed, from 1 July 2010 to 31 March 2013, already include a wide range of health measures, patient choice, tumour characteristics, demographics and hospital resource variables collected at diagnosis via pre-treatment patient interview and case note review. Follow-up of the cohort involved a further case note review (up to 3 years following diagnosis), postal survey (at 3–4 years post diagnosis – ideally all patients would be surveyed at 3 years, but timings are fixed by diagnosis dates of cohort) and mortality flagging (Figure 19).

FIGURE 2.

Kaplan–Meier breast cancer-specific survival curve for patients not treated with surgery vs. treated with surgery for breast cancer.

Only patients recruited/diagnosed from 1 July 2010 to 31 December 2012 are included in this study (n = 910) because recruitment was phased out in the final 6 months of the project, with the majority of sites stopping recruitment from 31 December 2012. Only another 34 out of the 944 patients were recruited from 31 December 2012 to 31 March 2013. The inclusion of these final 34 patients in this study is not necessary to support the analyses and would have increased study costs substantially, as it would have required a further wait of 3 months (with concomitant staff salaries, etc.) before the analysis could be conducted with only 34 more patients included (see Appendix 1).

We conducted further case note reviews to follow up each of the included 910 participants up to 3 years post diagnosis, recording adjuvant treatments received. The pro formas for collecting data from case notes were developed and piloted in consultation with clinicians and a health economist (to ensure the correct cost allocation for various procedures). Inter-rater reliability and data quality checks were undertaken on 10% of cases. Pro formas satisfied kappa > 0.6, showing substantial to perfect agreement and data input errors of < 0.3%. 11

The HRQoL survey undertaken at diagnosis was repeated. Ideally, this would have been at 3 years in all women, but timings are fixed to diagnosis date and the first woman was recruited into the study on 1 July 2010. As participants have already consented to further follow-up, ethics approval only required a substantial amendment to specify the follow-up instruments, etc. Patients who did not return the survey were followed up by telephone and reposting 2 weeks later and offered telephone support or a face-to-face interview to complete the survey.

The Office for National Statistics mortality flagging via NHS Digital provided information on both date and cause of death. This enabled analyses of breast cancer survival, the primary outcome of interest.

Outcome/dependent variables

The following outcome measures were specified a priori:

• aim A (extent to which primary surgery increases survival, HRQoL and is effective)

• aim Bi (extent to which adjuvant treatment increases survival and HRQoL and is effective)

• survival to 3 years post diagnosis

• HRQoL at 3–4 years from diagnosis (see above and Appendices 5–7 for explanation of timings)

• QALYs at 3–4 years post diagnosis

• aim Bii (extent to which lack of adjuvant treatment can be explained by patient health and choice)

• receipt of radiotherapy in addition to primary surgery

• receipt of chemotherapy in addition to primary surgery.

Explanatory variables

Explanatory variables include measures of health, patient choice, tumour characteristics and demographic variables. Adjusting for these in the analyses enabled us to account for case mix, health status and preferences of older women. Health measures have been selected based on ease of administration, validity, reliability, acceptability to older people, availability of normative data and prediction of non-standard management and/or treatment outcomes. 3,11,15,16 Tumour characteristics have been selected on the basis of management guidelines including TNM stage, grade and steroid receptor status. Choice was determined using the CPS, which has been selected as a validated measure of patient choice. 6–8,14

Analyses

Results

Sample size calculation

We were required to screen 1000 patients to enrol enough women into the PLACE trial using perometer measurements, which allowed us to determine if BEA had a > 80% sensitive and a > 80% specific accuracy. Currently the specificity of arm swelling measured by perometer is 87% specific for subsequent lymphoedema at 18 months with a sensitivity of 54% (assessed from ALMANAC data). 34

Older age, increases in body mass index (BMI) and postoperative radiotherapy are claimed to increase lymphoedema development. 21–23,28–31 We built a multivariate model predicting lymphoedema from the following potential predictor variables: BMI, dominant limb, postoperative radiotherapy, previous sentinel node biopsy, cigarette smoking, weight gain and age. This allowed us to identify what factors, as well as early arm-volume changes or BEA, predict subsequent development of lymphoedema. Although we anticipated 1000 patients recruited by 24 months of the programme to allow us to build a multivariate model, delays to sites opening meant that 1100 were recruited by June 2015. Multiple logistic regression modelling techniques were used to identify significant predictors of lymphoedema at early (18 months) and late time points (24 months) in the participants.

Self-reported symptoms and quality-of-life measures

Patients were asked to complete a lymphoedema questionnaire, which used three items from the Lymphedema and Breast Cancer Questionnaire about heaviness, numbness and swelling, and the Functional Assessment of Cancer Therapy – Breast Cancer, version 4 (FACT-B+4) Questionnaire (www.facit.org/FACITOrg/Questionnaires) and the EuroQol-5 Dimensions (EQ-5D) (www.euroqol.org/about-eq-5d.html) to assess self-reported upper limb symptoms, physical functioning disease-specific QoL and health utility. All questionnaires were completed preoperatively and then again at 3 and 6 months post surgery, with the exception of the EQ-5D, which was not completed at 3 months post surgery.

Statistical analysis

Statistical analysis included sensitivity and specificity analysis of the BIS L-Dex score against the ‘gold standards’ of perometer assessment at 6, 18 and 24 months (and subsequently clinical sleeve application) using statistical techniques recommended by Bland and Altman. 35 The BIS value cut-off level was checked using receiver operating characteristic (ROC) analysis and confirmed using later results. An assessment of the relationship between the two methods of measurement up to 2 years in predicting lymphoedema was performed. The analysis for the current report involved comparison of the baseline and 6-, 18- and 24-month post-surgery measurements using paired t-tests, and comparison between groups defined by lymphoedema status using independent t-tests, and data were described using means and ranges, sensitivity and specificity, and univariate and multivariate analyses. ROC analysis, Cox proportional hazards regression, log-rank testing and generalised estimating equation (GEE) regression were performed for univariate and multivariate analyses. The GEE regression was chosen as the inference was at the population level so the GEE marginal effects were of interest. Descriptive methods were used for all other data presented.

Results

Out of the 1100 patients entered into the study (median follow-up 36 months, minimum 24 months), the mean age was 55.7 years (range 22 to 90 years), 47.0% had a mastectomy and ANC, 90.5% were node positive, 70.9% had a histology of infiltrating ductal carcinoma and the majority (80.6%) were estrogen receptor (ER) positive (Table 8). Eighty-three per cent received postoperative radiotherapy, 67.3% received chemotherapy and 82.4% were given endocrine treatment. Fifty-eight patients (5%) had no post-1-month perometer measurements. Overall, 497 patients have completed 60 months’ follow-up, 105 have died and 204 have been lost to follow-up (or withdrawn from the study).

Lymphoedema assessment within the protocol was defined as the development of an arm-volume increase (RAVI) of > 10% but, on reviewing the data, compression sleeves were being applied outside the protocol indications. For some patients, significant swelling of the lower arm or hand or swelling of < 10% RAVI but associated with symptoms led to the application of a compression sleeve by the lymphoedema practitioner. We ascertained this by retrieving the perometer readings for all patients with a sleeve applied. Compression garment application was another surrogate marker of lymphoedema and potentially a better clinical marker. Time to lymphoedema for RAVI of > 10% at follow-up and that for sleeve application are presented.

The median time to developing lymphoedema was 11.3 months (range 2.3–63.1 months). Lymphoedema incidence (sleeve application and RAVI of > 10%) is shown in Tables 9 and 10. The incidence of lymphoedema differed by the definition of either a clinical sleeve application by a lymphoedema nurse or the perometer RAVI of > 10% after 24 months’ follow-up.

Using Kaplan–Meier estimates for time to diagnosis of lymphoedema, 14.6% using RAVI and 17.7% by sleeve application were diagnosed by 12 months, and 21.4% and 24.4% were diagnosed by 24 months, respectively (Figure 9).

There was clinical lymphoedema diagnosis/applied sleeve in 24.4% patients by 24 months, compared with 21.4% with RAVI of > 10% during follow-up. The majority of this difference appeared to occur between 6 and 12 months, when more sleeves were applied. This was partly due to the patients who were not eligible for the PLACE trial but had a perometer value RAVI of > 9% and who therefore went on to compression sleeves.

Lymphoedema by 24 months detected in 21.4% of women by perometry whereas using BIS definition in 39.4%. A moderate correlation between perometer and BIS at 6 months (r = 0.61) was found, with a sensitivity of 76% (95% CI 64% to 84%), specificity of 85% (95% CI 83% to 88%) and PPV of BIS of 31% (95% CI 25% to 39%) (Table 11). Sensitivity remained similar at 24 months (75%, 95% CI 64% to 83%), though specificity was higher (91%, 95% CI 89% to 93%), as was PPV of BIS (54%, 95% CI 44% to 63%). The sensitivity and specificity values for BIS fall below the percentage of 95% required according to the study protocol.

Women who developed a RAVI of > 5 to < 10% by 6 months required lymphoedema treatment in 35% of cases by 24 months, whereas a RAVI of < 3% was associated with an 8% lymphoedema rate at 24 months (p < 0.001).

The sensitivity and specificity of BIS and perometer were compared according to sleeve application using sleeve application as the clinical diagnosis of lymphoedema, as well as against the protocol defined perometer, RAVI of > 10%. There were 226 patients with an appropriately applied sleeve, which included patients with a sleeve applied and patients who had > 10% lymphoedema and were offered a sleeve by the lymphoedema nurse but declined it because they already had metastatic disease and were about to die. In addition, 51 patients had their sleeve applied as part of the PLACE trial; nine patients contralateral sleeve application and these patients were excluded from the analysis.

Perometer by 6 months

The perometer by 6 months variable includes those women with lymphoedema at 3 or 6 months.

For those with lymphoedema according to the RAVI of ≥ 10% definition, the BIS value used is the one at the time of the lymphoedema diagnosis. For those without lymphoedema according to the RAVI of ≥ 10% definition, the largest BIS value at either time point was used (Figure 3).

FIGURE 3.

Comparison of perometer and BIS at 6 months. Q1, RAVI only of ≥ 10; Q2, both ≥ 10; Q3, neither ≥ 10; Q4, BIS only of ≥ 10. Sensitivity, 69% (59/86; 95% CI 58% to 77%); specificity, 82% (698/851; 95% CI 79% to 84%); PPV, 28% (59/212; 95% CI 22% to 34%); NPV, 96% (698/725; 95% CI 95% to 97%). NPV, negative predictive value.

At all time points BIS identified high numbers of false-positive patients with a lymphoedema diagnosis (using RAVI of > 10%).

Relative arm-volume increase after 6 months up to 18 months

The RAVI after 6 months up to 18 months variable excludes those with lymphoedema up to and including 6 months.

For those with lymphoedema according to the RAVI of ≥ 10% definition, the BIS value used is the one at the time of the lymphoedema diagnosis. For those without lymphoedema according to the RAVI of ≥ 10% definition, the BIS value is the largest value from 9 to 18 months (Figure 4).

FIGURE 4.

Comparison of RAVI and BIS at 18 months. Q1, RAVI only of ≥ 10; Q2, both ≥ 10; Q3, neither ≥ 10; Q4, BIS only of ≥ 10. Sensitivity, 68% (53/78; 95% CI 57% to 77%); specificity, 81% (600/738; 95% CI 78% to 84%); PPV, 28% (53/191; 95% CI 22% to 34%); NPV, 96% (600/625; 95% CI 94% to 97%). NPV, negative predictive value.

Relative arm-volume increase of > 10% after 6 up to 24 months (excludes those with lymphoedema up to and including 6 months)

For those with lymphoedema according to the RAVI of ≥ 10% definition, the BIS value used is the one at the time of lymphoedema diagnosis. For those without lymphoedema, the BIS value used is the largest value between 9 and 24 months (Figure 5).

FIGURE 5.

Comparison of RAVI and BIS after 6 months up to 24 months. Q1, RAVI only of ≥ 10; Q2, both ≥ 10; Q3, neither ≥ 10; Q4, BIS only of ≥ 10. Sensitivity, 68% (15/22; 95% CI 58% to 76%); specificity, 79% (572/722; 95% CI 76% to 82%); PPV, 31% (68/218; 95% CI 25% to 38%); NPV, 95% (572/604; 95% CI 93% to 96%). NPV, negative predictive value.

Clinical lymphoedema/appropriately applied sleeve from 6 up to 18 months

The clinical lymphoedema/applied sleeve between 6 and 18 months variable excludes those with lymphoedema up to and including 6 months (see Table 12 and Figure 6).

FIGURE 6.

Comparison of RAVI of > 10% and BIS of > 10% with sleeve application. NB those with no clinical lymphoedema or applied sleeve by 6 months may have had an applied sleeve at a later time point.

At all time points, BIS would have resulted in more sleeves being applied than RAVI of > 10% incorrectly with lower specificity (Tables 12–15). RAVI of > 10% was better at identifying patients who could be reassured and did not need surveillance proving cost-effective to the NHS. It ruled out patients unlikely to develop lymphoedema and has a higher PPV at 18 months.

For those with lymphoedema, the BIS and perometer values used are those at the time of lymphoedema diagnosis. For those without lymphoedema, the BIS and perometer values used are the largest value between 9 and 18 months.

Up to 6 months: sleeve application in at least one-third of patients appeared to be as a result of a composite of significant arm symptoms (swelling and heaviness) together with arm-volume increases. The PPV for RAVI of > 10% at both time points is superior to BIS, although the negative predictive values (NPVs) are similar (Figure 7).

FIGURE 7.

Comparison of RAVI of > 10% and BIS with sleeve application (6–18 months). Note that those with no clinical lymphoedema or appropriately applied sleeve by 6 months may have had clinical lymphoedema or an applied sleeve at a later time point.

Clinical lymphoedema/applied sleeve after 6 months up to 24 months

The clinical lymphoedema/applied sleeve after 6 up to 24 months variable excludes those with lymphoedema up to and including 6 months.

For those with lymphoedema, the BIS and RAVI values used are those at the time of lymphoedema diagnosis. For those without lymphoedema, the BIS and RAVI values used are the largest value between 9 and 24 months (Table 16).

It is apparent that perometer is more specific (94–96%) than BIS (80–91%) at all time points. In other words, RAVI measurement gets more diagnoses of sleeve application correct and fewer wrong, particularly at 6 months. Thus, it is noticeable that at 6 months BIS of > 10 had 170 false positives, yet identified only 40 out of the 74 sleeves that were applied, whereas perometer identified 30 out of the 75 sleeves applied but overdiagnosed only 55 rather than 170 patients (Table 17). The differences in sleeve application numbers reflect the fact that some patients did not have a BIS measurement. BIS use would have meant that patients had far more sleeves applied than those using RAVI of > 10%, but neither method was particularly sensitive (see Figure 8).

FIGURE 8.

Comparison of RAVI of > 10% and BIS with sleeve application (6–24 months). Note that those women with no clinical lymphoedema or appropriately applied sleeve after 6 months and up to 24 months may have had clinical lymphoedema or a sleeve applied at a later time point.

We then looked at the sensitivity and specificity of RAVI and BIS at 6, 18 and 24 months (i.e. comparing both methods with each other). Once again, BIS of > 10 identified those patients with a RAVI of > 10% measurement in 68–76% of cases.

Combined relative arm-volume increase or bioimpedance spectroscopy versus clinical lymphoedema/appropriately applied sleeve

We considered whether combining RAVI of > 10% and BIS of > 10% improved the diagnosis of lymphoedema compared with sleeve application. At all time points it reduced PPV, although sensitivity increased slightly (Table 18).

Clinical lymphoedema/applied sleeve after 6 months up to 24 months (excludes those with lymphoedema up to and including 6 months)

The BIS and RAVI values used are those at the time of the indicated lymphoedema. For those without lymphoedema, the BIS and RAVI values used are the largest value between 9 and 24 months (Table 19).

The 85 patients with lymphoedema are made up of 39 with both RAVI and BIS of ≥ 10, 30 with only BIS of ≥ 10 and 16 with only RAVI of ≥ 10%.

Predictive value of bioimpedance spectroscopy value by 6 months against lymphoedema by 18 or 24 months

Lymphoedema defined by RAVI of ≥ 10% and clinical lymphoedema or applied sleeve.

In all of the analyses that follow, any patients diagnosed with a RAVI value of > 10% by 6 months were excluded from the analysis (n = 87) and any patients with a clinical lymphoedema or sleeve applied before 6 months were excluded from the analysis (Table 20).

There is a significant relationship between both BIS category by 6 months and lymphoedema defined by perometer of ≥ 10% by 18 months (p < 0.001) and clinical lymphoedema or applied sleeve by 18 months (p < 0.001).

For lymphoedema defined by RAVI of > 10% the significant relationship appears to be as a result of the higher rate of lymphoedema, 24%, in those with BIS score of ≥ 10 (which is the BIS definition of lymphoedema).

For clinical lymphoedema/applied sleeve there appears to be a small increase in lymphoedema rate across the BIS < 3, ≥ 3 to < 5, and ≥ 5 to < 10 categories from 7% to 16% across the three categories. The significant relationship appears mainly to be as a result of the higher rate of lymphoedema, 36%, in those with BIS of ≥ 10 (Table 21).

There is a relationship between both BIS category by 6 months and lymphoedema defined by perometer of ≥ 10% (p < 0.001) and clinical lymphoedema or applied sleeve by 24 months (p < 0.001).

For lymphoedema defined by perometer of ≥ 10% there appears to be little difference in the lymphoedema rate in the < 3, ≥ 3 to < 5, and ≥ 5 to < 10 categories; the rate was between 9% and 15% in each of those categories. The relationship appears to be as a result of the higher rate of lymphoedema, 30%, in those with BIS of ≥ 10.

For clinical lymphoedema or applied sleeve, there is an increase in the lymphoedema rate across all four categories, with smaller increases across the first three categories and a larger increase in the rate of lymphoedema in those with BIS of ≥ 10, which is the diagnostic category for lymphoedema according to BIS.

Prediction of lymphoedema

Two analyses were performed: one looked at the situation described above (lymphoedema after 6 months and up to 2 years), and the other looked at the time to first lymphoedema including all follow-up data (1-month visit was excluded as per the protocol, version 5.2, and the NIHR Programme Grants for Applied Research programme response letter). Both RAVI (> 10%) and sleeve application were considered in these analyses.

Factors predicting lymphoedema from baseline

For RAVI of > 10% as the end point, univariate analysis revealed BMI (p = 0.004), age (p = 0.013), previous sentinel node biopsy (p = 0.027), ER status (p = 0.006: ER negativity: HR 1.59, 95% CI 1.14 to 2.21), and number of nodes involved (p < 0.001) all predicted lymphoedema development. If one only considers those with confirmed or absent lymphoedema by 24 months, 25% (47/190) of ER-negative patients and 18% (141/794) of ER positive patients developed lymphoedema.

The multivariable analysis included number of nodes involved (HR 1.04, 95% CI 1.02 to 1.06), age ≥ 70 years compared with age < 70 years (HR 1.67, 95% CI 1.10 to 2.55), BMI of > 30 kg/m² (HR 1.62, 95% CI 1.14 to 2.32) and ER negativity (HR 1.56, 95% CI 1.11 to 2.19) in the model for predicting lymphoedema development (Table 22).

Sleeve application from baseline 24 months (excluding 1-month lymphoedema)

Univariate analysis revealed that only two factors predicted the sleeve application, node positivity (per-node increase) (HR 1.04, 95% CI 1.02 to 1.05) and adjuvant radiotherapy (HR 2.08, 95% CI 1.30 to 3.33), and both were independent in the multivariable analysis with a HR of 1.03 (95% CI 1.01 to 1.05; p < 0.001) and a HR of 1.93 (95% CI 1.20 to 3.10; p = 0.007), respectively (Table 23).

Lymphoedema development after 6 months’ surveillance (i.e. 6 months up to 2 years and the time to first lymphoedema within that time)

Patients with lymphoedema at 3 or 6 months are excluded because the inclusion of the RAVI variable, which is determined at 6 months, means there would need to be a ≥ 10% category RAVI variable but this is also used as the outcome event. In addition, excluding these patients is part of the study protocol (version 5.2) and the NIHR Programme Grants for Applied Research programme response letter.

The RAVI of ≥ 10% univariate analysis revealed that BMI (p < 0.002), number of nodes involved (median 2, range 0–41; p < 0.001), largest RAVI change by 6 months (p < 0.001; HR 5.58 for ≥ 5% to < 10% vs. < 3%, 95% CI 3.61 to 8.62) and BIS of > 10% (p < 0.001) all predicted lymphoedema development from 6 months up to 2 years.

The multivariable analysis included RAVI change by 6 months (p < 0.001; HR 5.22 for ≥ 5% to < 10%, 95% CI 3.22 to 8.47) along with number of nodes involved (HR 1.05, 95% CI 1.02 to 1.07), adjuvant chemotherapy (HR 1.61, 95% CI 1.01 to 2.55), BMI of > 30 kg/m² (HR 1.87, 95% CI 1.16 to 3.02) and BIS > 10% (p = 0.069) in the model for predicting lymphoedema development after 6 months up to 2 years (Table 24).

Smoking, type of surgery, weight gain and histological tumour type were not significant [n = 1100: those with lymphoedema ≤ 6 months have been excluded]. Factors predicting time to lymphoedema (sleeve applied) after 6 months surveillance (excluding lymphoedema at 6 months).

For applied sleeves as the clinical definition of lymphoedema, univariate analysis revealed that adjuvant radiotherapy (p = 0.008), adjuvant chemotherapy (p = 0.005), ER status (p = 0.076), ER negativity (HR 0.63, 95% CI 0.38 to 1.05), BIS of ≥ 10% (p < 0.001) and RAVI of ≥ 10% (p < 0.001) all predicted time to lymphoedema after 6 months to 24 months.

The multivariable analysis included adjuvant radiotherapy (p = 0.021), adjuvant chemotherapy (p = 0.003), ER status (p = 0.012), ER negativity (HR 0.51, 95% CI 0.30 to 0.86), BIS of ≥ 10% (p < 0.001) and perometer of ≥ 10% (p < 0.001) independently predicted time to lymphoedema after 6 months to 24 months (Table 25).

At 1 month, because both chemotherapy and radiotherapy had not commenced, the prediction was not as good as that at 6 months. Early changes in arm volume (RAVI) had the highest HR and were the best single predictor of subsequent lymphoedema.

Quality-of-life analyses: FACT-B scores

There are nine FACT-B+4 summary scores:27 physical well-being (PWB; score range 0–28), social/family well-being (SWB; score range 0–28), emotional well-being (EWB; score range 0–24), functional well-being (FWB; score range 0–28), breast cancer subscale (BCS; score range 0–40), arm subscale (ARM; score range 0–20), FACT-G total score (FACT-G = PWB + SWB + EWB + FWB; score range 0–108), FACT-B total score (FACT-B = PWB + SWB + EWB + FWB + BCS, score range 0–148), and Trial Outcome Index (TOI = PWB + FWB + BCS; score range 0–96). Descriptive summary data are shown in Appendix 16.

A simple comparison of the QoL data at each time point separately revealed that patients with lymphoedema at 6 months (by either definition) had significantly lower FACT-B+4, TOI and ARM subscale scores (Table 26 and 27). Poorer ARM subscale scores were also found at 12, 18 and 24 months (Table 28). At all time points, a significantly higher percentage of patients reporting swelling symptoms was found for those with lymphoedema, and a significantly higher percentage of patients reporting heaviness symptoms was found at 6, 18 and 24 months. Lower FACT-B+4, TOI and ARM subscale scores were found in the two smaller restricted subsets of patients defined by ‘no sleeve’ usage. These lower scores were significant for TOI (Table 27), and for the ARM subscale at 6, 12, 18 and 24 months. Likewise, a higher percentage of patients with reported swelling and patients with reported heaviness symptoms were found for those with lymphoedema in the smaller restricted subsets of patients defined by ‘no sleeve’ usage, and these remained significant (Table 29). The absolute change in FACT-B+4 (p = 0.04), TOI (p = 0.046) and ARM subscale (p = 0.009) scores between 6 and 24 months were significantly related to having lymphoedema at 24 months.

Generalised estimating equation regression analysis to further analyse the changes in quality-of-life scores over time

FACT-B Trial Outcome Index

Owing to the negative skew of the TOI variable, a transformation [log normalised (LN) (120 – TOI)] was used for the GEE analysis to obtain a better approximation to a normal distribution. In a regression model including the time variable, the estimated marginal mean (EMM) of TOI at each time point is presented in Table 30. A total of 997 patients had some data in the model. There was a change in TOI over time (p < 0.001).

TABLE 30 - FACT-B TOI estimated marginal mean at each time point

Time point	Estimated marginal mean of TOI	95% CI
Pre surgery	68.0	67.2 to 68.9
3 months	63.5	62.5 to 64.5
6 months	65.4	64.4 to 66.4
12 months	70.2	69.2 to 71.1
18 months	70.6	69.6 to 71.5
24 months	71.0	70.0 to 71.9

A GEE analysis that included an interaction term between lymphoedema status by 6 months and time showed that TOI varied over the time period (p = 0.003), those with lymphoedema by 6 months had significantly lower TOI overall (p = 0.028) and the interaction between time and lymphoedema status was significant (p < 0.001). There was a difference in the pattern of change over time between those with and those without lymphoedema (Table 31 and Figure 10).

The EMMs from the interaction term in the GEE analysis are presented below and in Table 31.

TABLE 31 - FACT-B TOI EMMs at each time point for those with and without lymphoedema by 6 months

The p-values in this table have not been adjusted for multiple testing.

Time point	Estimated marginal mean of TOI (95% CI)		p-value
	Without lymphoedema by 6 months (n = 883)	With lymphoedema by 6 months (n = 87)
Pre surgery	68.3 (67.4 to 69.2)	67.0 (63.9 to 70.0)	0.42
3 months	63.8 (62.7 to 64.8)	61.4 (57.8 to 64.8)	0.19
6 months	66.0 (64.9 to 67.1)	60.1 (56.4 to 63.5)	0.001
12 months	70.7 (69.8 to 71.7)	65.3 (61.8 to 68.5)	0.001
18 months	71.1 (70.2 to 72.1)	65.4 (61.9 to 68.6)	0.001
24 months	71.4 (70.4 to 72.4)	67.6 (64.0 to 71.0)	0.033

The main effect for the time variable was significant (p < 0.001). The main effect for the lymphoedema status by 6 months variable was significant (p = 0.028), showing that there was a difference between the lymphoedema status groups overall. Patients who developed lymphoedema by 6 months did not initially have poorer QoL (TOI) scores than those who did not develop lymphoedema, but by 6 months their scores were poorer and they remained poorer until 24 months, when the difference was no longer significant.

FIGURE 9.

The FACT-B TOI EMMs over time for all patients and those with and those without lymphoedema at 6 months.

It is noteworthy that those without lymphoedema at 6 months begin to regain their QoL (TOI) at 3 months, improving to be above pre-surgery levels by 12 months, whereas those who develop lymphoedema continue to have worsening QoL until 6 months and do not regain QoL to pre-surgery levels until 24 months, and do not surpass their pre-surgery QoL scores. This is an important finding, as it clearly and robustly indicates poorer QoL in the group who develop lymphoedema following surgery.

FACT-B total scores

Owing to the negative skew of the FACT-B total score variable, a transformation [LN(160 – FACT-B)] was used for the GEE analysis to obtain a better approximation to a normal distribution.

There was a change in FACT-B total over time (p < 0.001), and a borderline difference in the overall level of the FACT-B scores between those with and without lymphoedema (p = 0.055) over time (Figure 10). Although the difference between patients with and without lymphoedema indicated deficits in QoL from 3 months onwards, the pattern of change over time between groups was different (p < 0.001). The EMMs from the GEE analysis are in Table 32. By 6 months patients without lymphoedema regained their pre-surgery levels of total FACT-B score, whereas those who developed lymphoedema had persistent FACT-B QoL deficiency until 12 months after surgery (see Table 32 and Figure 10).

FIGURE 10.

The FACT-B total score EMMs over time for all patients and those with and without lymphoedema at 6 months.

TABLE 32 - The FACT-B total scores EMM at each time point for those with and without lymphoedema by 6 months

The p-values in this table have not been adjusted for multiple testing.

Time point	EMM of FACT-B total score (95% CI)		p-value
	Without lymphoedema by 6 months (n = 882)	With lymphoedema by 6 months (n = 87)
Pre surgery	110.9 (109.7 to 112.2)	109.4 (105.0 to 113.4)	0.48
3 months	107.8 (106.3 to 109.2)	105.3 (100.3 to 109.9)	0.32
6 months	110.1 (108.6 to 111.5)	104.0 (98.7 to 108.8)	0.018
12 months	115.5 (114.1 to 116.8)	109.5 (104.5 to 114.0)	0.012
18 months	115.9 (114.5 to 117.2)	109.2 (104.2 to 113.7)	0.005
24 months	116.0 (114.6 to 117.4)	111.8 (106.5 to 116.7)	0.11

ARM subscale

Owing to the negative skew of the ARM subscale variable, a transformation [LN(22 – ARM)] was used for the GEE analysis to obtain a better approximation to a normal distribution (data for 995 patients).

There was a change in scores over time (p < 0.001) for FACT-B ARM scores (Figure 11 and Table 34), and a difference in the overall level of the ARM scores between those with and without lymphoedema (p = 0.002). The pattern of change over time was different between the two groups (p < 0.001). For ARM subscale values, all patients’ values declined which did not return to baseline by 24 months implying a long-term arm symptom increase with ANC Surgery. ARM scores were persistently worse in those patients who developed lymphoedema, and there was a decrease in ARM subscale EMMs from pre surgery to 3 months in both those with and those without lymphoedema by 6 months. The EMM of those without lymphoedema by 6 months remained similar level to the EMM at 3 months before increasing slightly at 24 months but remained below the pre-surgery EMM. The EMM score at 24 months increased, although it remained below the pre-surgery EMM (Table 33).

FIGURE 11.

Change in FACT-B ARM scores EMMs for patients with and without lymphoedema at 6 months.

TABLE 33 - ARM subscale scores EMM at each time point for those with and without lymphoedema by 6 months

The p-values in this table have not been adjusted for multiple testing.

Time point	EMM of ARM subscale (95% CI)		p-value
	Without lymphoedema by 6 months (n = 882)	With lymphoedema by 6 months (n = 87)
Pre surgery	18.7 (18.6 to 18.9)	18.6 (17.9 to 19.1)	0.62
3 months	16.2 (15.9 to 16.4)	15.7 (14.7 to 16.5)	0.29
6 months	16.0 (15.8 to 16.3)	14.0 (13.0 to 14.8)	< 0.001
12 months	16.2 (15.9 to 16.4)	14.4 (13.4 to 15.3)	< 0.001
18 months	16.3 (16.0 to 16.5)	14.7 (13.5 to 15.6)	0.001
24 months	16.5 (16.3 to 16.8)	15.2 (14.2 to 16.0)	0.003

TABLE 34 - FACT-B+4: analysis of factors influencing the QoL scores

CT, chemotherapy; RT, radiotherapy.

Variable	Analysis
	Univariate		Multivariable (n = 683)
	EMM (95% CI)	p-value	EMM (95% CI)	p-value
Lymphoedema in first 12 months
No	116.9 (115.4 to 118.4)		114.3 (112.1 to 116.5)
Yes	109.9 (105.9 to 113.6)	< 0.001	106.4 (101.7 to 110.8)	< 0.001
Adjuvant CT
No	117.9 (115.4 to 120.2)	0.051	–	–
Yes	114.9 (113.1 to 116.6)		–
Adjuvant RT
No	116.6 (113.2 to 119.7)	0.60	–	–
Yes	115.6 (114.0 to 117.1)		–
BMI (kg/m2) at baseline
≤ 25	120.3 (118.1 to 122.4)	< 0.001	116.5 (113.4 to 119.3)	< 0.001
> 25 to ≤ 30	114.8 (112.4 to 117.1)		109.4 (105.7 to 112.9)
> 30	110.3 (107.1 to 113.2)		105.0 (100.8 to 108.9)
Type of surgery
ANC/other	118.4 (115.7 to 120.9)	0.012	112.5 (108.7 to 115.9)	0.063
WLE + ANC	116.8 (114.2 to 119.3)		110.9 (107.1 to 114.5)
Mastectomy + ANC	113.4 (111.2 to 115.6)		108.1 (104.8 to 111.2)
Smoking
Never	117.1 (115.3 to 118.8)	0.012	113.9 (111.5 to 116.3)	0.018
Ex-smoker	114.7 (112.0 to 117.2)		111.1 (107.9 to 114.2)
Current	109.1 (103.0 to 114.5)		106.2 (99.5 to 112.1)
Age (years)
< 50	113.8 (111.1 to 116.3)	0.058	107.8 (103.9 to 111.5)	0.002
≥ 50	116.7 (115.0 to 118.4)		113.1 (110.4 to 115.6)

Clinical lymphoedema/appropriately applied sleeve

Whether the clinical lymphoedema or applied sleeve variable is considered or the RAVI of > 10% definition of lymphoedema, the multivariable analysis results are similar. The only difference of note is that type of surgery is no longer associated with FACT-B at 12 months, although the direction of effect is the same as when the RAVI of > 10% definition of lymphoedema was included in the model.

In addition to lymphoedema, the analysis identified BMI, current smoking and older age, which had not previously been reported to influence FACT-B scores.

A similar analysis at 6 months found that chemotherapy significantly influenced scores, but the effect was lost in multivariate analysis by 12 months (see Table 34).

This is the first large prospective analysis of factors affecting QoL using FACT-B+4 in breast cancer patients and has identified several new factors, such as smoking and BMI, which influence QoL. BMI was also found to have important effects on QoL in the health economic analysis, strengthening the case for encouraging weight loss strategies after cancer diagnosis.

Relationship between the lymphoedema checklist variables at 6 months and changes in quality of life from baseline and relative arm-volume increase/bioimpedance spectroscopy from 1 month

The lymphoedema checklist is a patient self-reported symptom checklist; the constituent symptoms were compared with QoL changes in the study.

Greater reductions (from baseline to 6 months) in FACT-B, TOI and ARM subscale scores were found in those with swelling, heaviness (p < 0.001) and numbness (p < 0.035 FACT-B; p = 0.051 TOI; and p < 0.001 ARM).

For patients reporting swelling at 6 months, the reductions (from baseline to 6 months) in FACT-B (p < 0.001), TOI (p < 0.001) and ARM subscale (p < 0.001) were greater than for those not reporting swelling.

For patients reporting numbness at 6 months, the reductions (from baseline to 6 months) in FACT-B (p = 0.035), TOI (p = 0.051) and ARM subscale (p < 0.001) were greater than for those not reporting numbness.

For patients reporting heaviness at 6 months, the reductions (from baseline to 6 months) in FACT-B (p < 0.001), TOI (p < 0.001) and ARM subscale (p < 0.001) were greater than for those not reporting heaviness.

Greater increases (from 1 month to 6 months) in the exact RAVI value and exact BIS values were found in those with swelling (both p < 0.001), and in those with heaviness (RAVI p = 0.038 and exact BIS p = 0.004).

There were greater increases (from 6 to 24 months) in the exact RAVI value in those with swelling (p < 0.001), and in those with heaviness (RAVI p = 0.001) both during the time period and the responses at 24 months. No association was seen between 6 and 24 months with exact BIS values and swelling, numbness or heaviness.

In summary, RAVI of > 10% or changes in RAVI more closely related to patient-reported symptoms of arm swelling and heaviness throughout the BEA study.

Associations between changes in quality of life from baseline to 6 months and perometry/bioimpedance spectroscopy at 6 months

There was a negative association between changes in TOI and RAVI (r = –0.10; p = 0.024) and BIS (r = –0.14; p = 0.001) at 6 months. Larger reductions in QoL scores were found in patients who have had larger RAVI/BIS increases.

Associations between changes in quality of life from baseline to 6 months and changes in perometry/bioimpedance spectroscopy from 1 month to 6 months

There was a negative association between changes in TOI and changes in RAVI (r = –0.10; p = 0.024) and BIS (r = –0.14; p = 0.001), and between changes in FACT-B and changes in BIS (r = –0.11; p = 0.011). Thus, again greater reductions in QoL scores were found in those patients who had bigger increases in RAVI/BIS values.

Quality-of-life overview

These results demonstrate significant and persisting impact of lymphoedema on QoL following diagnosis and treatment for breast cancer. This is true for overall disease-specific HRQoL indicators reflected in the FACT-B TOI and total FACT-B+4 scores and is pronounced in specific symptoms associated with lymphoedema reflected in the ARM subscale scores. While overall QoL (TOI) does return to (or exceed) pre-surgery levels by 12 months for those without lymphoedema and 24 months for those with lymphoedema, the arm-specific measures indicate deficits for all ANC patients continuing until at least 24 months.

Most QoL studies in breast cancer have been cross-sectional and have not used instruments designed specifically for lymphoedema. Most reported reduced QoL with onset of lymphoedema. 23,26,27 The ALMANAC QoL study, reported by Fleissig et al. , found similar TOI and FACT-B+4 reductions in the ANC arm and considered them due to surgery. 26 Whereas 66% of their patients were node negative and the majority did not receive chemotherapy, in contrast in the BEA study, all were node positive and 66% received chemotherapy. There was a significant relationship between chemotherapy and QoL during the 6 months of treatment, but patients developing lymphoedema in the first 6 months had a greater reduction in QoL than those who did not develop lymphoedema. Whereas after the period during which chemotherapy was administered, QoL returned to pre-surgical levels by 12 months in patients who did not develop lymphoedema, the QoL deficit was prolonged after the development of lymphoedema until at least 24 months. Current smoking, BMI and age also affected QoL scores. Self-reported symptoms were associated with lymphoedema development but were not discriminatory predictors on their own. 24 Subjective symptoms such as heaviness and particularly ‘considerable’ swelling correlated with QoL deficits, RAVI and BIS increase and are probably clear and simple markers of adverse effects for many patients. 23,26,27 Such simple to complete self-reported measures could play an important part in clinical practice if widely adapted and combined with objective measures, and as seen below they may also contribute significantly to predicting the development of lymphoedema.

Composite scoring model to define lymphoedema

Cohort of patients who had clinical lymphoedema or appropriately applied sleeve only (excluding the PLACE trial patients)

A total of 223 patients had clinical lymphoedema (appropriately applied sleeve) by 24 months (79 by 6 months, 168 by 12 months); this excludes those who had a sleeve applied during the PLACE trial.

There were significant increases in the proportion of women with swelling, numbness and heaviness from the time point previous to sleeve fitting to the time point the sleeve was fitted. There was a decrease in the ARM subscale from the FACT-B+4 QoL questionnaire from before the sleeve was fitted to the time the sleeve was fitted (Table 36).

The rates of symptoms reported in our BEA study are higher than those reported in the ALMANAC trial. 26 The combination of arm swelling symptoms and nurses’ perception of poorer QoL of these women (as reflected by their QoL scores) may have led to the application of compression sleeves in these patients even though the RAVI was < 10% (Table 37). The perometer measurements for these patients are retained at each site and the arm-volume changes over the different segments were reviewed with the case note/source documents to understand the basis for sleeve application in the women for whom RAVI was < 10% in order to be able to produce a composite measure of lymphoedema.

We identified localised segmental swelling in the hand or upper, lower arm segments on source perometry measurements such that if the forearm segment had a 10% volume increase (even if the whole arm RAVI was < 10%), a compression sleeve was applied based on these clinical findings, and symptoms. 23

Even after central review of source perometry data, there remained patients for whom decisions regarding sleeve fitting were determined by lymphoedema nurses based on patient reports of worsening symptoms, rather than on objective measurement of arm swelling. This finding is in line with results of the qualitative study reported in WS3 below.

The RAVI and BIS values increased from before lymphoedema to the time of diagnosis but the mean values for RAVI post sleeve application were < 9%, which implies that, in women with subthreshold arm-volume increases whose symptoms worsened, sleeves were used to treat symptoms in the absence of objective volume criteria defining lymphoedema.

Changes in quality of life in relation to sleeve application

Overall, for patients who required treatment with a sleeve, an increase in their QoL occurred. Comparing FACT-B+4 at two time points – the time the sleeve was applied and approximately 6 months after – there was a mean increase of 2.96 (p = 0.021: t-test). Using repeated measures [analyses of variance (ANOVAs) shown in Table 43], total FACT-B (p = 0.015), ARM (p < 0.001) and TOI (p < 0.001) all showed that QoL decreased from baseline to the point before their sleeve was applied. FACT-B and TOI showed an improvement at the time at which the sleeve was applied. At 24 months, FACT-B and TOI returned to above pre-surgery levels, but the ARM subscale remained low.

Arm symptoms were not improved as much as overall QoL after sleeve application.

Repeated measures ANOVAs showing the changes in different QoL measurements at various points in relation to the time of sleeve application (Table 38).

The patients split into two groups with regard to QoL changes following sleeve application. One group were patients who met the conventional definition of lymphoedema (having a RAVI of ≥ 9% at sleeve application) and another group contained patients who had a RAVI of < 9% at sleeve application. Patients with complete data sets including (both their RAVI and total FACT-B+4) before and after sleeve application were analysed. Sixty patients with sleeves applied were not included because of missing data.

The patients who had a larger increase in arm volume (median RAVI was 11.7%) had a mean FACT-B score of 106.3 when their sleeve was applied. From sleeve application to the first time point afterwards, approximately 6 months after, their QoL showed a large increase to 112.6 (p = 0.004), suggesting that reducing arm swelling by treating lymphoedema is an important factor in improving their QoL.

The group with smaller amounts of arm swelling (i.e. with a median RAVI of 3.6%) had a lower mean FACT-B score of 103.5 when their sleeve was applied, which increased by a small amount, from 105.5 (p = 0.20). However, the arm volume in this group continued to increase, suggesting that the sleeve was not an effective treatment.

Effect of self-reported arm swelling on quality-of-life benefit following sleeve application

Within FACT-B+4 there are five questions, which relate directly to lymphoedema, including B3, which relates to arms being either swollen or tender.

Repeated measures ANOVAs of total FACT-B, TOI and ARM at the time of sleeve application and afterwards, split by the patients’ B3 score at the time of application B3 scores, are reverse coded, so a score of 0 is ‘very much’ arm swelling and 4 is ‘not at all’.

The initial overall QoL scores for patients with little or no swelling are, as expected, significantly higher than those for patients with considerable self-reported arm swelling. Likewise, patients with little arm swelling have higher QoL scores (FACT-B+4, TOI) at 36 months post surgery.

Patients were grouped based on their self-reported B3 scores at the time of sleeve application. The group with large amounts of arm swelling had B3 scores of 0–2 (81% of those with RAVI of ≥ 9% had a B3 score of 0–2 at sleeve application) and the group with little to no arm swelling had B3 scores of 3–4 (54% of those with RAVI of < 9% had a B3 score of 3–4) (p ≤ 0.005).

ARM subscale scores increased (improved QoL) when sleeve was applied for ‘considerable’ arm swelling (B3 score 0–2) but were unaltered when little or no swelling was present (Figure 13).

FIGURE 13.

Change in the ARM subscale score at the time of sleeve application and afterwards, by B3 score.

In the ARM subscale (p = 0.044) analysis, there was an interaction between having a B3 score of 0–2 (more arm swelling) or 3–4 (limited arm swelling) and the time points post sleeve application. The interaction term was significant because of the increase in ARM scores after sleeve application in patients with B3 scores 0–2, whereas there was minimal change in the ARM score post sleeve application in patients with a B3 score of 3–4.

In the patients with considerable swelling, mean FACT-B and TOI increased (QoL improved) following sleeve application. Among the group with B3 scores showing limited or no arm swelling, there were small QoL increases. Patients with small amounts of arm swelling had higher QoL scores throughout until 36 months after their surgery (Table 39).

These data indicate that when patients self-report ‘considerable’ arm swelling, application of a sleeve in ‘correctly diagnosed’ lymphoedema successfully improves symptoms and QoL scores. However, if a sleeve is applied without self-reported arm swelling and/or with no RAVI of > 9% (definition of lymphoedema), no benefit in arm symptoms or QoL occurs.

The prescription of compression sleeves in ‘correctly diagnosed’ lymphoedema successfully improves symptoms and QoL. Understanding and developing objective evidence for which patient groups benefit from treatment with a compression sleeve has important implications for compression sleeve prescription and use in the NHS.

Scoring model to predict lymphoedema

The definition of lymphoedema used as the outcome for the logistic regression included both RAVI of > 10% or a sleeve applied after 1 or 6 months up to 24 months.

Of the 1097 patients in the data set, 326 were classified as having either an appropriately applied sleeve or clinical lymphoedema.

Fifty-one patients were identified as being given their sleeve as part of the PLACE trial and nine had a sleeve applied to the contralateral arm (because of deep-vein thrombosis). These 60 patients were excluded from consideration in the following analysis.

There were 266 patients with an appropriately applied sleeve or clinical lymphoedema.

Model at 6 months predicting lymphoedema (relative arm-volume increase of > 10%)

The variables considered for the scoring model were RAVI at 6 months (categorical), BIS at 6 months (categorical), TOI at 6 months, FACT-B total at 6 months, ARM subscale at 6 months, lymphoedema checklist questions at 6 months (swelling, numbness, heaviness), B3 at 6 months (categorical: 0–2, considerable swelling vs. 3–4, little to no swelling), age, BMI at 6 months, ER status, number of positive nodes, adjuvant chemotherapy and adjuvant radiotherapy.

A total of 711 patients were included in this analysis.

A scoring model was produced based on the regression coefficients from the final model (Table 40). The individual scores are the regression coefficients for binary or categorical variables rounded to the nearest 0.5 and the regression coefficients for continuous variables to two decimal places owing to their per-unit increase interpretation. The total ‘diagnostic’ score is given by summing the individual scores. A patient with a higher total score is more likely to have a 10% RAVI volume increase.

This scoring model gives an area under the receiver operating characteristic (AUROC) of 0.80 (95% CI 0.74 to 0.85). For a cut-off score of 1.58 – where a patient with a score of ≥ 1.58 would be predicted to have a 10% RAVI by perometer – the scoring model would give a sensitivity of 80.0% (68/85), a specificity of 67.7% (424/626), a PPV of 25.2% (68/270) and a NPV of 96.1% (424/441). The cut-off score was chosen to maximise the sum of the sensitivity and specificity, giving equal weight to both.

Prediction scoring predicting lymphoedema (relative arm-volume increase of > 10%) at 6 months (excluding bioimpedance spectroscopy)

Bioimpedance spectroscopy is not widely available in the NHS and therefore we concentrated on models that could be used in any lymphoedema clinic in the UK. A total of 740 patients were included in this analysis (Table 41).

A scoring model was produced based on the regression coefficients from the final model as described previously. The total ‘diagnostic’ score is given by summing the individual scores. A patient with a higher score is more likely to have a 10% RAVI perometer volume increase.

This scoring model gives an AUROC of 0.77 (95% CI 0.71 to 0.82). For a cut-off score of 1.41, where a patient with a score of ≥ 1.41 is predicted to have a 10% perometer volume increase, the scoring model would give a sensitivity of 72.1% (62/86), a specificity of 72.2% (472/654), a PPV of 25.4% (62/244) and a NPV of 95.2% (472/496). The cut-off score was chosen to maximise the sum of the sensitivity and specificity, giving equal weight to both. This model has similar AUROC and would be easy to apply in NHS practice and its components would have few extra costs (i.e. FACT-B+4/lymphoedema checklist) in any NHS setting.

Model of prediction of lymphoedema (relative arm-volume increase of > 10%) from 1 month

Variables considered for the scoring model were RAVI at 1 month (categorical), BIS at 1 month (categorical), TOI at pre-surgery, FACT-B total at pre-surgery, ARM subscale at pre-surgery, lymphoedema checklist questions at pre-surgery (swelling, numbness, heaviness), B3 at pre-surgery (categorical: 0–2, considerable swelling vs. 3–4, little to no swelling), age, BMI at pre-surgery, ER status, number of positive nodes, adjuvant chemotherapy and radiotherapy.

A total of 522 patients were included in this analysis.

A scoring model was produced based on the regression coefficients from the final model as described previously (Table 42). The total ‘diagnostic’ score is given by summing the individual scores. A patient with a higher score is more likely to have a 10% RAVI perometer volume increase.

This scoring model gives an AUROC of 0.71 (95% CI 0.64 to 0.77). For a cut-off score of 0.82, where a patient with a score of ≥ 0.82 is predicted to have a 10% perometer volume increase, the scoring model would give a sensitivity of 62.9% (56/89), specificity of 70.7% (306/433), PPV of 30.6% (56/183) and NPV of 90.3% (306/339). The cut-off score was chosen to maximise the sum of the sensitivity and specificity, giving equal weight to both.

Prediction model: using sleeve as ‘lymphoedema’ definition after 6 months

The variables considered for the scoring model were perometer at 6 months (categorical), BIS at 6 months (categorical), TOI at 6 months, FACT-B total at 6 months, ARM subscale at 6 months, lymphoedema checklist questions at 6 months (swelling, numbness, heaviness), B3 at 6 months (categorical: 0–2, considerable swelling vs. 3–4, little to no swelling), age, BMI at 6 months, ER status, number of positive nodes, adjuvant chemotherapy and adjuvant radiotherapy (Table 43).

Patients with a RAVI of ≥ 10% before, or at, 6 months were excluded from the analysis.

A total of 548 patients were included in this analysis.

A scoring model was produced based on the regression coefficients from the final model as described previously. The total ‘diagnostic’ score is given by summing the individual scores. A patient with a higher total score is more likely to have a lymphoedema requiring a sleeve.

This scoring model gives an AUROC of 0.76 (95% CI 0.70 to 0.82). For a cut-off score of 4, where a patient with a score of ≥ 4 is predicted to have clinical lymphoedema or a sleeve applied, the scoring model would give a sensitivity of 48.6% (34/70), a specificity of 90.0% (430/478), a PPV of 41.5% (34/82) and a NPV of 92.3% (430/466). The cut-off score was chosen to maximise the sum of the sensitivity and specificity, giving equal weight to both.

Again, this model uses simple measures easily available in the NHS and provides good prediction of lymphoedema.

Model predicting lymphoedema (sleeve) development from 1 month post surgery

The variables considered for the scoring model were: perometer at 1 month (categorical), BIS at 1 month (categorical), TOI at pre-surgery, FACT-B total at pre-surgery, ARM subscale at pre-surgery, lymphoedema checklist questions at pre-surgery (swelling, numbness, heaviness), age, BMI at pre-surgery, ER status, number of positive nodes, adjuvant chemotherapy and adjuvant radiotherapy (Table 44).

A total of 837 patients were included in this analysis.

A scoring model was produced based on the regression coefficients from the final model. The individual scores are the regression coefficients for binary or categorical variables rounded to the nearest 0.5 and the regression coefficients for continuous variables to two decimal places due to their per-unit increase interpretation. The total ‘diagnostic’ score is given by summing the individual scores. A patient with a higher total score is more likely to have a clinical lymphoedema or require a sleeve.

This scoring model gives an area under the receiver operator characteristic (AUROC) of 0.67 (95% CI 0.62 to 0.71).

Note that the ARM subscale is significant if included in the above model. However, only 558 patients would be included in the model and the area under the curve (AUC) is not improved by a large amount by its inclusion (AUC 0.67, 95% CI 0.62 to 0.73).

These models provide reasonable prediction of patients at risk of lymphoedema and those for 1 month after surgery have only three variables and are simple to apply in a clinical setting.

Models from 6 months have higher AUROC and are a better fit because patients have completed their treatments at that point.

Health economics: estimation of health-related utility measures – data available for analysis

A data extract was created in July 2017 containing EuroQol-5 Dimensions, three-level version (EQ-5D-3L) data collected from 1100 BEA patients for up to 24 months.

Successful completion of all of the five dimensional questions (i.e. mobility, self-care, usual activities, pain/discomfort and anxiety/depression) required for calculation of the utility measure was disappointing, declining steadily from 82% to 62% during the 2-year period from baseline (Table 45). Most of the incomplete records contained no data for any of the five questions and were therefore unusable for analysis. Completed documents were more often missing for obese patients (BMI of > 30 kg/m²: p < 0.0001) and current smokers (p < 0.004).

The response to the EuroQol visual analogue scale (EQ VAS) question (requiring only a simple cross on a scale between 0 and 100) was similarly disappointing (Table 46).

Of particular interest is the number of patients supplying a continuous sequence of complete EQ-5D-3L data from baseline onwards, as this allows temporal changes in estimated health-related utility to be tracked over time, and correlated with clinical events and the development of lymphoedma. As many as possible of the EQ-5D ratings spoiled by missing dimension entries were remedied by tracing similarities in the pattern of response in earlier and later completed forms, and interpolating where the patient showed consistency of response over time. EuroQol forms with missing responses, which could not be remedied by imputation, were excluded from subsequent analyses.

Table 47 shows that a full complete EQ-5D-3L record after imputation of missing values was available for only 37% of the patient sample, and for 17% of patients no data were provided at all.

Data imputation

Most of the 40 EQ-5D ratings spoiled by missing dimension entries were remedied by tracing similarities in the pattern of response in earlier and later completed forms and interpolating where the patient showed consistency of response over time. Only four were found to be wholly or partly irredeemably flawed and excluded from subsequent analyses.

Data analysis

The objective of the analysis carried out on this data set was to identify and quantify the mean change in the EQ-5D-3L utility estimate attributable to the presence of confirmed clinical lymphoedema. Ideally, this would be carried out by using complete sequences of utility estimates over 24 months, and comparing those recorded for patients developing lymphoedema with those for patient who remained lymphoedema-free throughout. Unfortunately, the poor completion rates described above result in only 140 patients developing lymphoedema during the trial and also having a full valid sequence of EQ-5D-3L responses from baseline to 24 months after imputation: limiting the reliability of estimates of disutility obtained. However, alternative methods of analysis have been explored in order to obtain an approximation to the magnitude of the effect of lymphoedema on patient experience.

A search for potentially confounding patient characteristics likely to affect the estimation of patient utility values identified a prospective cohort study of lymphoedema patients at the University of Pennsylvania Lymphoedema Clinic, which described 124 patients with upper extremity cancer lymphoedema and reported EQ-5D-3L results. 42 The severity of the lymphoedema had little effect on the mean estimated utility value, but the authors reported strong associations between estimated utility and BMI, higher BMI being associated with lower utility scores.

An initial exploratory regression analysis of the BEA data confirmed a similarly strong association between patient BMI and EQ-5D-3L utility estimates in our study.

Figure 14 demonstrates that obese and very obese patients are more likely to develop lymphoedema (chi-squared test, p = 0.023). As patient recruitment to the BEA study was not randomised, it was necessary in any comparison between subcohorts to apply a corrective adjustment to counter baseline differences in BMI.

FIGURE 14.

Association between the incidence of lymphoedema (assessed by perometry) and patient BMI.

Another patient characteristic known to influence patient-reported utility is the age of patients. An analysis was undertaken of the relationship between the age at which patients entered the BEA study, and their propensity to develop lymphoedema in the 2 years following surgery.

Figure 15 shows that differences in the distribution of patients by age is less pronounced between those who did and did not develop lymphoedema during the study, which is confirmed by a non-significant chi-squared test result (p = 0.39). Correlation analysis between age and utility estimates confirmed that no significant bias is associated with variations by age. Therefore, it was concluded that no adjustment for age was necessary to standardise between the two cohorts.

FIGURE 15.

Association between the incidence of lymphoedema (assessed by perometry) and patient age.

For a total of 403 patients, full EQ-5D responses were submitted or derived by imputation across the 2-year period from baseline (i.e. five data sets at 6-monthly intervals) and a valid BMI could be calculated. Of these, 137 (34%) were identified with primary lymphoedema during the 2-year period, and 266 (66%) were lymphoedema-free throughout. The mean EQ-5D utility estimates are shown in Figure 16, unadjusted for BMI. Patients with lymphoedema are shown in three subgroups according to whether the diagnosis was made by perometry, by clinical assessment with fitting of a compression sleeve, or both in combination.

FIGURE 16.

Mean utility estimates at 6-monthly intervals, by method of assessment leading to diagnosis of lymphoedema, unadjusted for BMI differences.

Figure 17 shows the same comparison following adjustment of utility estimates in the three lymphoedema subgroups to match the mean BMI in the lymphoedema-free group at individual patient level to a common BMI average (26.6 kg/m²). It is important to note that none of the graphical differences in either chart is statistically significant, because of the small number of cases in each of the lymphoedema subgroups.

Nonetheless, it is possible to identify suggestive patterns in these data:

• There is a consistent loss of estimated patient utility at the 6-month assessment relative to the preoperative (baseline) values, consistent with the impact of recovering from surgery.

• By the 12-month assessment, there is a general recovery of at least some of the initial utility loss.

• Patients who do develop lymphoedema in the 24-month period post surgery appear to recover to similar utility levels to those recorded at baseline.

• The subgroups which featured the need for a compression sleeve to be fitted when lymphoedema was diagnosed (whether or not perometry was used to confirm the diagnosis) generally failed to recover to preoperative utility levels during follow-up.

• The very small subgroup who were found to have developed lymphoedema by perometry but were not deemed clinically to require a sleeve fitting appear to recover to preoperative utility levels (or possibly better), although this may be related to small number uncertainty.

FIGURE 17.

Mean utility estimates at 6-monthly intervals, adjusted for BMI differences at individual patient level.

Of particular interest in assessing the cost-effectiveness interventions aimed at reducing the incidence and/or impact of lymphoedema is the estimation of patient-reported disutility attributable to experiencing lymphoedema over an extended period of time. This involves comparing utility estimates for patients with and patients without lymphoedema from the BEA study population.

This has been performed by identifying any patient with a diagnosis of lymphoedema by any method within the first few months after baseline, for whom a full sequence of five utility estimates are available. This limits consideration to patients with at least 6 months experience of the condition. A total of 41 patients fulfilled this criterion, and their utility values at 12, 18 and 24 months were compared with data at the same time points from patients never experiencing lymphoedema.

The results are shown in Table 48 and indicate consistent non-zero disutility estimates at all three time points.

Averaged over this 12-month period of observation, patients with extended experience of lymphoedema recorded a mean utility score of 0.721, compared with 0.823 for patients with no recorded lymphoedema, giving an estimated mean disutility attributable to lymphoedema of –0.102 (95% CI –0.127 to –0.076). If BMI-adjusted utility values are used instead, the size of this effect would reduce to –0.073 (95% CI –0.122 to –0.023).

A search of the literature for comparable research-based estimates of the disutility associated with lymphoedema following breast surgery proved fruitless. Only one published cost-effectiveness study included an assumed value for disutility of –0.03, justified only as the ‘smallest clinically important difference in utility’. 2,43 The estimates obtained using the available BEA data, although not definitive, are statistically significant and evidence-based, and should, therefore, be considered superior.

Further analysis of the BEA data will be possible and will also be performed for PLACE trial participants.

Data analysis: lymphoedema incidence

Another important statistic required to carry out a cost-effectiveness analysis is the incidence rate of the key outcome variable, in this case the proportion of patients confirmed to suffer from clinical lymphoedema, and the timing of such events.

Data on the first recorded time of confirmed lymphoedema have become available for a period exceeding 5 years from baseline. This has made a Kaplan–Meier analysis of the timing of the first recorded lymphoedema event (i.e. the duration of the initial lymphoedema-free period from baseline) possible, as displayed in Figure 18.

FIGURE 18.

Kaplan–Meier survival analysis of primary lymphoedema incidence in BEA patients, compared with the Pennsylvania study. 44

This exhibits a typical profile as seen in studies where assessments are carried out at predetermined intervals, but that the precise timing is spread over several weeks around the target time. This gives rise to periods of time between planned assessments when only a very few ‘opportunistic’ primary lymphoedema events are recorded, followed by multiple events occurring either side of each planned assessment time. To mitigate the bias introduced by the study design, it is necessary to identify an ‘envelope’ of accurate data points corresponding to the time at the end of ‘step-down’ phase of the prespecified assessment times. At these points all planned and opportunistic events up to that time are included. These ‘envelope’ data are represented in the chart by the large circles in Figure 18.

Using sections of the envelope data, it was possible to estimate the annual incidence rate of primary lymphoedema at different periods of time. In the first 6 months from baseline, the incidence rate was 16.4% per annum, contrasting with 9.5% between 9 and 24 months, and 4.9% between 36 and 60 months. Clearly, the risk of lymphoedema-free patients suffering a primary event decreased steadily throughout the 5-year observation period.

Figure 18 also shows comparable 5-year lymphoedema incidence from a study of 631 breast cancer patients in Philadelphia and Delaware Counties, Pennsylvania. 45 Although these patients suffered higher incidence of lymphoedema throughout, the difference between the two trends is wholly attributable to a much higher incidence in the first 12 months (74% lymphoedema-free in the Pennsylvania study, compared with 85% lymphoedema-free in the current study), but the risk of developing lymphoedema thereafter was very similar to that found in the current study.

To estimate the incidence of primary lymphoedema at future times beyond the available data set, a range of standard statistical parametric models was fitted to the envelope data. However, this proved disappointing, with poor correspondence to the trial data when some functions were tested and unrealistic estimates of the mean long-term time spent lymphoedema-free for other functions.

An alternative approach was attempted, which sought to incorporate the existence of an unknown proportion of the study population who were at zero risk of lymphoedema. This method did not generally improve the correspondence of fitted models to the study data and led to a wide variation in the estimates of the zero-risk subset of the population (between 33% and 58%). Therefore, it has been concluded that without additional evidence from other sources it is not possible to obtain reliable estimates of the number of patients suffering lymphoedema beyond the available data and the timing of incident events.

Workstream 1

The consumer panels of three cancer research networks were consulted in the development of the research protocol for WS1. The response was positive, supportive and constructive in all cases. For example, one panel wrote, ‘This is a most excellent study that is badly needed’. Recommendations from the panels were integrated into the design of the study.

Similarly, the suggestions of patient representatives on the cancer research networks were followed for training research staff. Patient representatives consented to carry out training interviews with research staff giving them feedback on their interview technique. This proved particularly useful to provide insight into and experience of interacting with supportive ‘patients’ themselves. Recommendations about the design from both WS1 and WS2/3 panels were integrated into the design and helped with the Trial Management Group.

Workstream 2

A group of 10 people with lymphoedema from University Hospital of South Manchester were involved in helping us to develop better treatments for lymphoedema, primarily with regard to different designs of compression sleeves. We involved them in the design of this trial and the quality-of-life measures. They commented that they would have preferred to have had earlier intervention with external compression garments than to have undergone manual lymphatic drainage and compression therapy once they had developed lymphoedema.

Two patients who developed lymphoedema within 2 years of ANC surgery agreed to sit on the patient management group and were both initially involved with this project. Unfortunately, both died during the first 3 years of the project, and two further patient representatives were subsequently involved. A qualitative study consulted patients to understand the poor recruitment in the PLACE trial.

We involved the PPI forum within University Hospital of South Manchester to ensure that patients and the public were formally involved in plans to carry out research, monitor progress, implement findings and monitor the impact on services. We have liaised with charities that have a role in patient support, including BCC and Breakthrough Breast Cancer, to involve the public and patients in this research.

What was and was not successful in the Programme Grant

The programme of work in elderly breast cancer involving multicentre studies successfully recruited and provided important work on the management of elderly breast cancer. Workstream 2 recruited well after ANC and has produced clear results about the value of arm volume measurements and the use of, and indications for, compression arm sleeves. A health economics analysis was less successful because patients who were acutely affected by their cancer diagnosis and morbidity were reluctant to fill in QoL questionnaires. The trial of compression garments in patients developing early arm swelling failed to recruit sufficient patients because of lack of equipoise among lymphoedema nurses and clinicians. Thus, results of the PLACE trial are still awaited.

Implications for practice

Our findings suggest that older women should be offered surgery, which can be performed under local anaesthetic block if there are concerns over fitness, and surgeons need to make clear the advantages of surgical excision of the cancer on breast cancer survival. The lymphoedema prediction index described will aid communication and individualisation of monitoring of patients after ANC surgery. The use of the Lymphoedema Checklist in women after ANC surgery will aid early recognition of arm problems, particularly if it proves as good a marker of need for intervention as arm measurements. BIS does not reach the expected sensitivity or specificity compared with perometry and the Lymphoedema Checklist to justify its cost and introduction to NHS practice [this was the subject of a recent NICE Medical Technologies Evaluation Programme review www.valueinhealthjournal.com/article/S1098-3015(17)3246-9/fulltext]. Sleeves are not effective in the absence of RAVI of > 9% or ‘considerable’ self-reported arm swelling.

Model at 6 months

The variables considered for the scoring model were: perometer at 6 months (categorical), BIS at 6 months (categorical), TOI at 6 months, FACT-B total at 6 months, ARM subscale at 6 months, lymphoedema checklist questions at 6 months (swelling, numbness, heaviness), B3 at 6 months (categorical: 0–2, considerable swelling vs. 3–4, little to no swelling), age, BMI at 6 months, ER status, number of positive nodes, adjuvant chemotherapy and adjuvant radiotherapy.

Prediction scoring model 1

A total of 711 patients were included in this analysis.

A scoring model was produced based on the regression coefficients from the final model. The individual scores are the regression coefficients for binary or categorical variables rounded to the nearest 0.5 and the regression coefficients for continuous variables to 2 decimal places due to their per-unit increase interpretation. The total ‘diagnostic’ score is given by summing the individual scores. A patient with a higher total score is more likely to have a 10% perometer volume increase.

This scoring model gives an AUROC of 0.80 (95% CI 0.74–0.85) (Figure 24).

FIGURE 24.

Prediction scoring model 1: AUROC curve. Diagonal segments are produced by ties.

For a cut-off score of 1.58, where a patient with a score of ≥ 1.58 would be predicted to have a 10% perometer volume increase, the scoring model would give a sensitivity of 80.0% (68/85), specificity of 67.7% (424/626), PPV of 25.2% (68/270) and NPV of 96.1% (424/441). The cut-off score was chosen to maximise the sum of the sensitivity and specificity, giving equal weight to sensitivity and specificity.

Prediction scoring model 2: excluding bioimpedance spectroscopy at 6 months

A total of 740 patients were included in this analysis.

A scoring model was produced based on the regression coefficients from the final model as described previously. The total ‘diagnostic’ score is given by summing the individual scores. A patient with a higher total score is more likely to have a 10% perometer volume increase.

This scoring model gives an AUROC of 0.77 (95% CI 0.71 to 0.82) (Figure 25).

FIGURE 25.

Prediction scoring model 2: AUROC curve. Diagonal segments are produced by ties.

For a cut-off score of 1.41, where a patient with a score of 1.41 or above would be predicted to have a 10% perometer volume increase, the scoring model would give a sensitivity of 72.1% (62/86), a specificity of 72.2% (472/654), a PPV of 25.4% (62/244) and a NPV of 95.2% (472/496). The cut-off score was chosen to maximise the sum of the sensitivity and specificity, giving equal weight to sensitivity and specificity.

Prediction scoring model at 1 month

The variables considered for the scoring model were perometer at 1 month (categorical), BIS at 1 month (categorical), TOI at pre-surgery, FACT-B total at pre-surgery, ARM subscale at pre-surgery, lymphoedema checklist questions at pre-surgery (swelling, numbness, heaviness), age, BMI at pre-surgery, ER status, number of positive nodes, adjuvant chemotherapy and adjuvant radiotherapy.

Prediction scoring model 3

A total of 506 patients were included in this analysis.

A scoring model was produced based on the regression coefficients from the final model as described previously. The total ‘diagnostic’ score is given by summing the individual scores. A patient with a higher total score is more likely to have a 10% perometer volume increase.

This scoring model gives an AUROC of 0.71 (95% CI 0.65 to 0.77) (Figure 26).

FIGURE 26.

Prediction scoring model 3: AUROC curve. Diagonal segments are produced by ties.

For a cut-off score of 1.25, where a patient with a score of ≥ 1.25 would be predicted to have a 10% perometer volume increase, the scoring model would give a sensitivity of 65.5% (55/84), specificity of 69.9% (295/422), PPV of 30.2% (55/182), and NPV of 91.0% (295/324). The cut-off score was chosen to maximise the sum of the sensitivity and specificity, giving equal weight to sensitivity and specificity.

Prediction scoring model 4: excluding bioimpedance spectroscopy at 1 month

A total of 522 patients were included in this analysis.

A scoring model was produced based on the regression coefficients from the final model as described previously. The total ‘diagnostic’ score is given by summing the individual scores. A patient with a higher total score is more likely to have a 10% perometer volume increase.

This scoring model gives an AUROC of 0.71 (95% CI 0.64 to 0.77) (Figure 27).

FIGURE 27.

Clinical lymphoedema prediction scoring model at 6 months. Diagonal segments are produced by ties.

For a cut-off score of 0.82, where a patient with a score of ≥ 0.82 would be predicted to have a 10% perometer volume increase, the scoring model would give a sensitivity of 62.9% (56/89), specificity of 70.7% (306/433), PPV of 30.6% (56/183), and NPV of 90.3% (306/339). The cut-off score was chosen to maximise the sum of the sensitivity and specificity, giving equal weight to sensitivity and specificity.

Prediction scoring model at 6 months

The variables considered for the scoring model were perometer at 6 months (categorical), BIS at 6 months (categorical), TOI at 6 months, FACT-B total at 6 months, ARM subscale at 6 months, lymphoedema checklist questions at 6 months (swelling, numbness, heaviness), B3 at 6 months (categorical: 0–2, considerable swelling vs. 3–4, little to no swelling), age, BMI at 6 months, ER status, number of positive nodes, adjuvant chemotherapy and adjuvant radiotherapy.

Patients who had a perometer value ≥ 10 before or at 6 months were excluded from the analysis.

A total of 528 patients were included in this analysis.

Prediction model at 6 months

A scoring model was produced based on the regression coefficients from the final model as described previously. The total ‘diagnostic’ score is given by summing the individual scores. A patient with a higher total score is more likely to have a clinical lymphoedema or require a sleeve.

Prediction scoring model at 6 months

This scoring model gives an AUROC of 0.78 (95% CI 0.72 to 0.84) (Figure 28).

FIGURE 28.

Prediction scoring model 2: AUROC curve. Diagonal segments are produced by ties.

For a cut-off score of 4, where a patient with a score of 4 or above would be predicted to have clinical lymphoedema or a sleeve applied, the scoring model would give a sensitivity of 59.4% (41/69), specificity of 80.4% (369/459), PPV of 31.3% (41/131) and NPV of 92.9% (369/397). The cut-off score was chosen to maximise the sum of the sensitivity and specificity, giving equal weight to sensitivity and specificity.

Prediction scoring model: excluding bioimpedance spectroscopy at 6 months

A total of 548 patients were included in this analysis.

Prediction Model –excluding BIS at 6 months.

A scoring model was produced based on the regression coefficients from the final model as described previously. The total ‘diagnostic’ score is given by summing the individual scores. A patient with a higher total score is more likely to have a clinical lymphoedema or require a sleeve.

Prediction scoring model: excluding bioimpedance spectroscopy at 6 months

This scoring model gives an AUROC of 0.76 (95% CI 0.70 to 0.82) (Figure 29).

FIGURE 29.

Prediction of clinical lymphoedema excluding BIS at 6 months. Diagonal segments are produced by ties.

For a cut-off score of 4, where a patient with a score of ≥ 4 would be predicted to have clinical lymphoedema or a sleeve applied, the scoring model would give a sensitivity of 48.6% (34/70), specificity of 90.0% (430/478), PPV of 41.5% (34/82) and NPV of 92.3% (430/466). The cut-off score was chosen to maximise the sum of the sensitivity and specificity, giving equal weight to sensitivity and specificity.

Prediction scoring model at 1 month

The variables considered for the scoring model were: perometer at 1 month (categorical), BIS at 1 month (categorical), TOI at pre-surgery, FACT-B total at pre-surgery, ARM subscale at pre-surgery, lymphoedema checklist questions at pre-surgery (swelling, numbness, heaviness), B3 at pre-surgery (categorical: 0–2, considerable swelling vs. 3–4, little to no swelling), age, BMI at pre-surgery, ER status, number of positive nodes, adjuvant chemotherapy and adjuvant radiotherapy.

A total of 794 patients were included in this analysis.

Prediction model at 1 month

A scoring model was produced based on the regression coefficients from the final model. The individual scores are the regression coefficients for binary or categorical variables rounded to the nearest 0.5 and the regression coefficients for continuous variables to 2 decimal places due to their per-unit increase interpretation. The total ‘diagnostic’ score is given by summing the individual scores. A patient with a higher total score is more likely to have a clinical lymphoedema or require a sleeve.

Prediction scoring model at 1 month

This scoring model gives an AUROC of 0.67 (95% CI 0.63 to 0.72) (Figure 30).

FIGURE 30.

Prediction scoring model of clinical lymphoedema at 1 month. Diagonal segments are produced by ties.

For a cut-off score of 1.55, where a patient with a score of ≥ 1.55 would be predicted to have clinical lymphoedema or a sleeve applied, the scoring model would give a sensitivity of 47.1% (82/174), specificity of 79.8% (495/620), PPV of 39.6% (82/207), and NPV of 84.3% (495/587). The cut-off score was chosen to maximise the sum of the sensitivity and specificity, giving equal weight to sensitivity and specificity.

Prediction scoring model: excluding bioimpedance spectroscopy at 1 month

A total of 837 patients were included in this analysis.

A scoring model was produced based on the regression coefficients from the final model. The individual scores are the regression coefficients for binary or categorical variables rounded to the nearest 0.5 and the regression coefficients for continuous variables to 2 decimal places due to their per unit increase interpretation. The total ‘diagnostic’ score is given by summing the individual scores. A patient with a higher total score is more likely to have a clinical lymphoedema or require a sleeve.

Note: the ARM subscale is statistically significant if included in the above model. However, only 558 patients would be included in the model and the AUC is not improved by a large amount by its inclusion (AUC 0.67, 95% CI 0.62 to 0.73).

This scoring model gives an AUROC of 0.67 (95% CI 0.62 to 0.71) (Figure 31).

FIGURE 31.

Prediction scoring model of clinical lymphoedema at 1 month excluding BIS. Diagonal segments are produced by ties.

For a cut-off score of 1.55, where a patient with a score of ≥ 1.55 would be predicted to have a clinically identified lymphoedema or sleeve applied, the scoring model would give a sensitivity of 53.6% (98/183), specificity of 70.9% (464/654), PPV of 34.0% (98/288), and NPV of 84.5% (464/549). The cut-off score was chosen to maximise the sum of the sensitivity and specificity, giving equal weight to sensitivity and specificity.