Effectiveness and cost-effectiveness of traditional and new partner notification technologies for curable sexually transmitted infections: observational study, systematic reviews and mathematical modelling

Partner notification technologies requiring partner assessment at health services

In the UK, patient referral is the most commonly used partner notification technology for curable STIs. 14 When the number of cases of gonorrhoea increased markedly in the 1960s and 1970s, patient referral was introduced as an efficient way of bringing sexual partners to treatment. 15 Before this, contact tracing field staff usually notified the partners of people who had either syphilis or gonorrhoea (provider referral). Patient referral was then extended as a method of partner notification for chlamydia, which has since become the most commonly diagnosed STI in the UK.

A major advantage of patient referral, and other methods for which the partner has to attend the clinic, is that tests for other STIs, including HIV, can be done and additional prevention education can be offered at the same visit. The methods used to enable patient referral are now very diverse and can be separated into simple and enhanced categories, according to the intensity of the intervention (see Table 1).

Enhancements to improve the effectiveness of simple patient referral include additional written information, verbal advice and counselling, home-sampling kits for partners, and text messaging or websites to facilitate communication. 16–19

A health-care professional (the provider) can notify partners if the index patient does not want to or cannot do it him or herself. In practice, contract referral is often not seen as a separate method but as an extension of patient referral (if partners have not been notified by the time that a follow-up visit occurs), or as a prelude to provider referral if the index case wants to try first to inform a partner on his or her own. In the UK, advisers in sexual health (health advisers) are the specialist staff for partner notification and they usually work in genitourinary medicine (sexual health, STI) clinics. If the index case agrees that provider referral is acceptable for a particular partner, the health adviser then contacts the partner by telephone, post, e-mail, text message or, less frequently, by home visit. These approaches are most appropriate in instances when attendance of the partner can be confirmed, for example where a single clinic serves a geographic area. They are often used for partner notification for people with syphilis and blood-borne virus infections, including hepatitis B and C.

Partner notification technologies not requiring partner assessment at health services

Accelerated partner therapy (APT) is a new partner notification technology in the UK to provide quicker access to antibiotic treatment for sexual partners to reduce the time between index case diagnosis and partner treatment. 20 APT is based on expedited partner therapy (EPT), which was developed in the USA to increase efficient use of partner notification resources for the most common STIs, chlamydia and gonorrhoea. 21 EPT and APT can be considered as a specific type of enhanced patient referral;22 the enhancement is the provision of antibiotics or prescriptions for index cases to give to partners.

Expedited partner therapy removes the requirement for consultation with a health professional by allowing the index patient to give medication directly as a treatment package (patient-delivered partner therapy) or as a prescription to their partner(s). 21 In the UK, two models that comply with prescribing regulations have been developed: in APT Hotline a health adviser or senior nurse practitioner assesses the eligibility of the partner for treatment by telephone; in APT Pharmacy the sexual partner attends a community pharmacy and is assessed by a pharmacist. 20

The spread of undiagnosed and untreated STIs is the main concern regarding APT and EPT. 21 When antibiotics are provided without a clinic-based consultation, the sexual partner is not tested for STIs at the time of treatment and will not be advised on how to reduce the risk of future infection. If an index case diagnosed with chlamydia gives antibiotics for chlamydia alone to his or her sexual partner(s) and no tests for STIs are done, partners who are co-infected with gonorrhoea or HIV will not have these infections diagnosed. There is then a risk that untreated gonococcal and HIV infections could cause outbreaks, which would have been detected and contained if the partners had been tested. APT consultations therefore involve an ‘assertive invitation’20 to encourage partners to take antibiotics and to undergo testing for STIs, including HIV. Missed opportunities for testing and sexual health counselling might, however, be outweighed by the benefit of interrupting transmission of the index-case STI by reaching sexual partners who would otherwise have remained untreated. 23

Clinical effectiveness outcomes

Direct, objective outcomes of the primary end points of partner notification are the preferred measures of effectiveness. Primary end points are often not collected; however, because they are less frequent and more difficult to collect than intermediate outcomes. Reinfection of the index case by an untreated partner is now the preferred outcome of a failure of partner notification. The availability of nucleic acid amplification tests for diagnosis on non-invasively collected specimens has facilitated this. In practice, ‘reinfection’ is measured as repeated detection of infection at follow-up, and misclassification of the source of the repeated infection can occur. Without the additional use of highly discriminatory gene sequencing methods, repeated infection owing to antibiotic treatment failure or infection from a new partner cannot be excluded. 28 In addition, results can be biased if there is differential loss to follow-up between intervention arms.

Biological outcomes of pelvic inflammatory disease (PID) in women and population STI prevalence are difficult to attribute specifically to the impact of a partner notification intervention. To our knowledge, a reduction in the population prevalence of gonorrhoea and chlamydia is the stated primary outcome of one trial of a cluster-randomised trial of EPT in Washington (state), USA, but results have not yet been published. 29 Mathematical modelling studies are therefore the only way to estimate the impact of partner notification on transmission at the population level (see Mathematical modelling for decision-making about sexually transmitted infection control).

Intermediate outcome measures of the clinical effectiveness of partner notification include the number of partners of an index case that have been treated. In the UK, the number of partners treated per index case is the outcome used to monitor partner notification outcomes in health service evaluations, such as clinical audit. 30 Intermediate outcomes are often used as surrogates for successful partner notification but the relationship between the number of partners treated per index case and prevention of reinfection is not known. First, a record that a partner received treatment does not guarantee that medication was taken at all (unless directly observed). Second, the treatment has to be present in high enough concentrations to interrupt replication. Third, sexual partners have to avoid unprotected sexual intercourse for long enough to allow the treatment to work. For the sexual partner, the prevention of onward infection transmission is an important outcome. This can be defined as the number of transmissions that might have occurred from index cases’ partners to their sexual partners. 9 Secondary transmission is difficult to measure empirically in clinical studies because information about subsequent partners is often not collected and, if a secondary partner is infected, the direction of transmission is not known. It can, however, be estimated in modelling studies. In this monograph (see Chapter 2, Estimating the likely public health impact of partner notification for a clinical service: an evidence-based algorithm) we examine two new intermediate measures of the impact of partner notification, based on the concept of the ‘number needed to treat’: the absolute reduction in onward transmission (AROT) and its corollary, the number needed to interrupt transmission (NNTIT). 9

Cost-effectiveness outcomes

The UK National Institute for Health and Care Excellence (NICE) prefers the cost per quality-adjusted life-year (QALY) as the primary outcome for cost-effectiveness. The challenges in measuring clinical effectiveness in partner notification studies therefore also affect the estimation of cost-effectiveness. If the results of economic evaluations are to be presented in terms of cost per QALY, a valuation exercise is necessary. Economic evaluations of STI control interventions such as chlamydia screening programmes include the costs of preventing cases of PID, chronic pelvic pain, ectopic pregnancy and tubal factor infertility. These outcomes are therefore also relevant to evaluating partner notification as a health technology. Measuring health-related quality of life (HRQL) for patients during an episode of chlamydia infection presents several challenges. First, the asymptomatic nature of many STIs means that infections are often undiagnosed, and thus there is no apparent detrimental impact on quality of life. Second, instruments for measuring quality of life usually assume that a health state is chronic or permanent. This is not the case for acute episodes of STIs and for some sequelae such as PID and ectopic pregnancy. Infertility caused by a STI can be considered a permanent health state if a woman intended to start a family but a temporary health state if she decides not to have children. Cost per major outcome averted is often presented when utility weights cannot be determined. Presentation in natural units, however, cannot be directly compared with the cost-effectiveness of other interventions.

Intermediate economic outcomes of partner notification include the cost per new infection treated or of secondary cases prevented. 31 They are intermediate outcomes because they do not give a direct indication of the final outcome. These outcomes can be estimated in natural units (cost per case of chlamydia infection averted), but not in terms of health utilities. In addition, the cost per secondary transmission prevented cannot be measured directly and has to be estimated using transmission models of the infection.

Clinical effectiveness of partner notification

The studies in this section address objective (i) by examining the effects of both traditional and new partner notification technologies on different measures of clinical effectiveness. First, we compare primary outcomes of the clinical effectiveness of new technologies (EPT or APT) with traditional partner notification methods (simple and enhanced patient referral) for chlamydia, gonorrhoea and trichomonas, based on a systematic review and meta-analysis of randomised controlled trials. 44 Second, we reanalyse clinical audit data to estimate current levels of numbers of partners treated (intermediate outcome) of partner notification for chlamydia as achieved in UK genitourinary medicine clinics. 45 Third, we estimate the impact on secondary transmission of traditional methods of partner notification for different types of sexual partner seen in UK genitourinary medicine clinics, using static modelling. 9

Mathematical modelling of the impact of partner notification technologies

Studies in this section address objective (ii). First, we describe the modelling studies that we conducted to address uncertainties in estimates of the duration46 and transmissibility47 of C. trachomatis and to resolve differences between the results of previous individual-based models of C. trachomatis transmission. 48 We focus initially on chlamydia because it is the STI for which partner notification is now most often done. The roll-out of the NCSP in England means that partner notification has to be modelled as part of a screening intervention. We then describe how we developed new individual-based models to examine the impact of traditional partner notification methods, depending on whether the outcome is measured at the individual level or the population level49 (Table 5). Finally, we examine the effects of traditional (patient referral) and new (APT) technologies among heterosexuals, using a deterministic model. We developed a new model that allows the investigations of single infections with either chlamydia or gonorrhoea and co-infections to be examined, so that the effect of providing treatment for one STI without testing for the other could be explored.

Cost-effectiveness of partner notification

Studies in this section address objective 2. First, we examine the evidence available for obtaining QALYs for female reproductive tract outcomes of bacterial STIs for use in cost-effectiveness studies based on the primary population level outcomes of partner notification. Second, we used a static model to examine the intermediate outcomes of cost per case and per secondary case for current traditional partner notification technologies. 31

Discussion and conclusions

This section synthesises the findings from the component studies and provides implications for health care and recommendations for further research.

Introduction

Randomised controlled trials are the least biased study design for measuring the causal effect of an intervention on an outcome. Systematic reviews of randomised controlled trials about a specific research question use explicit protocols to identify, collate and synthesise evidence. This allows results to be combined statistically, where appropriate, and reasons for heterogeneous results between different trials to be explored. Previous systematic reviews of the relative effectiveness of different partner notification strategies for curable STIs did not find strong evidence for the superiority of simple patient referral, provider referral or contract referral, based on the intermediate outcomes that have been measured. 50 EPT has been developed as a new partner notification technology during the era of evidence-based medicine and has been evaluated in several large randomised controlled trials with repeat infection as a primary outcome. 16,51–53 EPT involves the use of antibiotic treatment packages or prescriptions as enhancements to patient referral for infected patients with curable STIs. For a comprehensive view of the role of EPT as a new partner notification strategy, it is relevant to describe comparisons between other forms of enhanced patient referral and simple patient referral, provider referral and contract referral. The most up-to-date evidence of the effectiveness of these technologies comes from the Cochrane Sexually Transmitted Infections Collaborative Review Group’s 2012 update of its systematic review of strategies for partner notification. 44 This chapter reports the findings relevant to the STIs, partner notification technologies and outcomes considered in this monograph. 44

Objective

To report the findings of randomised controlled trials comparing at least one method of enhanced patient referral with another partner notification technology in adults with curable STIs, evaluated using a biological outcome.

Methods

The protocol and methods are available in full from the Cochrane Database of Systematic Reviews. 44 In summary, the populations studied were patients with a diagnosis of chlamydia, gonorrhoea, non-specific genital infection, trichomonas, PID, syphilis or co-infection with any of these STIs. The interventions were strategies that aimed to enhance the effectiveness of patient referral using methods including health education and counselling, health education materials (such as pamphlets, posters, and video and audio productions), and patient assistance strategies directed at facilitating patient referral (such as referral cards, incentives, reminders, and video and audio presentations). EPT (including APT) was considered as a separate type of enhanced patient referral. Eligible comparisons were any other partner notification technology. The primary outcome was reinfection of the index case, measured as repeated detection at a follow-up visit. Secondary outcomes were the numbers of partners per index case who were notified, presented for care or treated and harms of partner notification.

Results

The full review in The Cochrane Library reports the results of the search strategy in detail. 44 Twenty-one randomised controlled trials16–19,50–65 reported on eight comparisons of a method of enhanced patient referral compared with an alternative technology in patients with curable STIs (Table 6). The updated search (5 January 2011 to 29 January 2012) identified 15 of these trials (9393 participants)16–19,51–57,59,61,64,65 and six (5331 participants)50,58,60,62,63 were included in the original review. Most trials (12) were conducted in the USA,16,17,50–54,56,58,60–62 three were done in the UK,19,55,64 two in Denmark,63,65 and one each in Australia,18 South Africa,50 Uganda57 and Zimbabwe. 59 There were no randomised controlled trials that evaluated the new UK partner notification technology of APT and no trials with reinfection as an outcome for syphilis, PID or for any STIs in men who have sex with men. There are no published results available for the cluster-randomised trial of EPT for chlamydia and gonorrhoea. 29

The results reported in the rest of this section are restricted to nine randomised controlled trials in which reinfection of the index case was either a primary or a secondary outcome. 16–19,51–55 Intermediate outcomes in these trials, including the number of partners treated, are also reported.

The trials could be grouped into three comparisons: EPT versus simple patient referral,16,19,51–54 EPT versus enhanced patient referral,16,19,53 and enhanced patient referral versus simple patient referral. 16,18,19,53,55 The methods used to enhance patient referral were EPT;16,19,51–54 use of booklets with tear-out information slips for index cases to give to their partner(s);16,53 additional counselling sessions;54 patient referral by a nurse at the time of receiving results at a general practitioners’ surgery rather than referral to a genitourinary medicine clinic; and use of a website. 18

The numbers of partners treated per index case randomised varied across trials and interventions (Table 7). This outcome was not reported in four17,18,52,53 of the nine trials. The median was 0.57 partners per index case (range 0.28–1.14). There was no clear relationship between the numbers of partners treated and the percentage of index cases with infection detected at the follow-up visit.

There was a risk of bias in at least one domain in all included studies. Random sequence generation was adequate in six16–19,53,55 of the nine trials that reported reinfection as an outcome. In three trials,51,52,54 random sequence generation was unclear. The method for allocation concealment was adequate in three trials. 18,52,55 In five trials,16,19,51,53,54 the method of allocation concealment was not sufficiently specified and in one17 the method could have introduced a high risk of bias; Wilson et al. 17 reported that participants were assigned study identification numbers sequentially as they enrolled in the study. Blinding of participants and personnel was not possible in any of the trials. Explicit blinding of laboratory personnel was reported in only one trial. 19 The attrition rate was > 20% in four16,19,51,54 of the seven trials16,19,51,54,56,59,61 in which repeat infection was the primary outcome and > 50% in both trials18,55 where repeat infection was measured as a secondary outcome.

Expedited partner therapy versus simple patient referral

Six studies (n = 6018) compared the rate of index patient reinfection EPT with simple patient referral among patients with chlamydia,19,52 trichomoniasis,53,54 or gonorrhoea or chlamydia. 16,51 The treatment pack for partners in all trials included antibiotics, written information about the infection, a telephone number for the study nurse and drug safety information. In addition, Cameron et al. 19 included information about genitourinary medicine clinics as alternative possibilities for seeking testing and treatment, Golden et al. 51 included condoms, and Schillinger et al. 52 included instructions for the index case to inform his or her partner about exposure to a STI and to abstain from sexual intercourse for 7 days after treatment.

The pooled results of trials for all infections showed that index patients in the EPT group had a 29% lower risk of being reinfected compared with index patients in the simple patient referral group (RR 0.71, 95% CI 0.56 to 0.89; heterogeneity p = 0.15, I² = 39%) (Figure 3).

FIGURE 3.

Forest plot of randomised controlled trials of EPT vs. simple patient referral, by infection and overall. PR, patient referral. Source: Reproduced from Ferreira A, Young T, Mathews C, Zunza M, Low N. Strategies for partner notification for sexually transmitted infections, including HIV. Cochrane Database Syst Rev 2013; 10:CD002843 http://dx.doi.org/10.1002/14651858/CD002843.pub2 with permission from John Wiley and Sons. 44 Copyright © 2013 The Cochrane Collaboration. Published by John Wiley & Sons, Ltd.

There was a modest level of between-trial heterogeneity, despite marked differences in the infections studied and follow-up testing interval. The size of the treatment effect appeared smaller in trials that included only women with chlamydia (RR 0.90, 95% CI 0.60 to 1.35) than in trials that included patients with either gonorrhoea or chlamydia (RR 0.61, 95% CI 0.39 to 0.94).

There were inconsistent findings about the effects of EPT interventions on intermediate outcomes. Three studies16,19,51 (n = 3600) assessed the number of partners notified. In one study,16 slightly more partners of index patients in the EPT group were notified (MD 0.45, 95% CI 0.28 to 0.62). In the other two studies19,51 the differences included the null effect (Golden et al. :51 MD –0.05, 95% CI –0.12 to 0.01; and Cameron et al. :19 MD 0.13, 95% CI –0.06 to 0.32).

Three16,51,54 studies assessed the number of partners treated. The studies showed results in the same direction but were very heterogeneous (heterogeneity p < 0.001, I² = 95%). In two16,54 of the three trials, there was a moderate difference favouring EPT (Kissinger et al. :16 MD 0.43, 95% CI 0.28 to 0.58; and Schwebke et al. :54 MD 0.51, 95% CI 0.35 to 0.67). The difference between groups was very small in the fourth trial (MD 0.06, 95% CI 0.01 to 0.12). 51

Expedited partner therapy versus enhanced patient referral

There were three trials16,19,53 that compared the effect on reinfection in the index case of EPT with enhanced patient referral among women with chlamydia19 or trichomonas,53 and men with chlamydial or gonococcal urethritis. 16 There was no evidence of a difference in the reinfection rate between the two groups (RR 0.96, 95% CI 0.6 to 1.53; heterogeneity p = 0.22, I² = 33%) (Figure 4).

FIGURE 4.

Forest plot of randomised controlled trials of EPT vs. enhanced patient referral. PR, patient referral. Reproduced from Ferreira A, Young T, Mathews C, Zunza M, Low N. Strategies for partner notification for sexually transmitted infections, including HIV. Cochrane Database Syst Rev 2013; 10:CD002843 http://dx.doi.org/10.1002/14651858/CD002843.pub2 with permission from John Wiley and Sons. 44 Copyright © 2013 The Cochrane Collaboration. Published by John Wiley & Sons, Ltd.

There was no consistent evidence of differences between the strategies in intermediate outcomes. Cameron et al. 19 (n = 220) found no evidence of a difference between groups in the number of partners presenting for care (MD 0.01, 95% CI – 0.02 to 0.03). Another study16 (n = 692) found a small increase in the number of partners treated per index patient randomised to the EPT group compared with the enhanced patient referral group (MD 0.22, 95% CI 0.21 to 0.23).

Discussion

The updated Cochrane systematic review shows that three additional randomised controlled trials comparing EPT with simple patient referral and two trials comparing EPT with enhanced patient referral have been published since an earlier review of EPT trials. 22 When data for all curable STIs are pooled, EPT results in a lower risk of reinfection in the index case when compared with simple patient referral, but not when compared with enhanced patient referral methods.

The strengths of this systematic review are the rigorous and reproducible methods used to search and appraise the literature. It is unlikely that the review missed trials that would change the conclusions. The weaknesses of the review relate mainly to the small number of studies with differences between interventions and reporting. This limits the ability to investigate differences in effectiveness for individual STIs and for specific types of enhanced patient referral. It was also not possible to draw firm conclusions about the relative effectiveness of EPT, enhanced patient referral and simple patient referral. Although EPT was superior to simple patient referral, and there was no statistical evidence of a difference between EPT and enhanced patient referral, there were too few comparable trials to estimate the direct comparison between enhanced and simple patient referral. A meta-analysis of indirect comparison would be useful to investigate this further. A further limitation results from measurement error in the primary outcome. Repeated detection of infection at a follow-up visit includes reinfection from an untreated partner but also persistent infection resulting from treatment failure and newly acquired infections. Misclassification should be non-differential, however.

The findings from the review of trials using primary biological outcomes cannot be extrapolated to syphilis or viral STIs and cannot be generalised to men who have sex with men. EPT has not been recommended as a partner notification method for men who have sex with men. 21

There were no randomised controlled trials evaluating the effectiveness of APT. Cameron et al. 19 evaluated an EPT intervention in Scotland by giving the index cases antibiotics to give to their partners. They noted that this was possible only in a research setting. APT has been described as a new model of EPT,21 but it is not clear whether or not the results of trials evaluating EPT, as developed in the USA, can be extrapolated to APT. The requirement for an assessment by a health-care professional in the APT Hotline and APT Pharmacy models might affect the potential uptake when compared with EPT. The finding that EPT can reduce the risk of reinfection in the index case in comparison with simple patient referral, but not enhanced patient referral, confirms that of our earlier systematic review. 22 It is not known whether or not there was a specific effect of EPT, for example on reducing the time to partner treatment, because information about treatment delays was not reported in any of the included trials. In a non-randomised evaluation of APT in the UK, Estcourt et al. 20 found that the numbers of days between the diagnosis of the index case and treatment of the sexual partner was slightly shorter (median 1 day) when APT was used than for routine partner notification by patient referral (median 3 days; p = 0.11). These results cannot necessarily be extrapolated to the randomised controlled trials in the Cochrane review. First, the routine partner notification described by Estcourt et al. 20 would be classified in the Cochrane review as enhanced patient referral because the consultation with a health adviser was supported with condoms, written information about the infection and advice for the partner to seek treatment. Second, the need for a partner assessment by telephone or pharmacist in the APT protocol might increase time to treatment, although the median time to treatment for EPT could only be reduced to 0 days.

The effects of treatment delays on the effects of APT compared with traditional patient referral are investigated in a modelling study (see Chapter 3, The effects of traditional and new partner notification technologies for C. trachomatis and N. gonorrhoeae).

Introduction

Data about the outcomes of partner notification, as achieved in routine clinical practice, can be used to inform input parameters for the uptake of partner notification in mathematical modelling studies. The levels measured in randomised controlled trials represent those achievable under controlled conditions and vary according to setting, infection, and type of partner notification technology (see Table 7). The British Association for Sexual Health and HIV carried out a national audit of case note documentation of chlamydial infection management in UK genitourinary medicine clinics in 2007. 66–68 These data allow auditable outcomes to be analysed and sources of variation to be examined. Biological outcomes, such as the rate of repeated detection, cannot be measured because repeat testing after treatment of positive cases is not included in treatment guidelines. 10 Similar audit data have previously shown marked differences in partner notification outcomes in genitourinary medicine clinics inside and outside Greater London. 30 This chapter reports data published by Herzog et al. 45

Objectives

To compare different ways of measuring outcomes for partner notification with published standards; to examine variability between clinics; and to examine factors at the individual and clinic levels that contribute to variation in partner notification outcomes.

Methods

We analysed data extracted from case notes of patients seen between 2 January 2007 and 31 March 2007 and responses about clinic-level policies relating to partner notification. Descriptions of the audit methods and responses have been published. 66–68

Results

We merged data on 5032 individuals with chlamydia in 193 clinics with clinic-level data reported by 177 genitourinary medicine clinics. The merged data set consisted of 4616 individuals in 169 genitourinary medicine clinics; this excluded 23 clinics with 415 index cases for which there was no information about clinic policies, and one clinic with only one index case for whom there were no data for any outcome of interest. The demographic characteristics of excluded index cases were comparable with those of included patients (results not shown). The percentages of patients for whom there was no information about the outcome of partner notification were partners tested for chlamydia, 41%; partners with a positive test for chlamydia, 43%; and partners treated for chlamydia, 32%.

Variability between clinics in partner notification outcomes

Figure 5 shows the mean number of partners tested for chlamydia per index case for each clinic, when health-care worker-verified data were supplemented with patient-reported values. The median across all clinics was 0.47 when missing values were assumed to be 0 (0.30 for Greater London genitourinary medicine clinics, 0.52 for all other clinics). No genitourinary medicine clinics were below the lower control limit but most Greater London clinics were below the median. Only one clinic was above the upper control limit, suggesting that most variation resulted from common causes. When index cases with missing data were excluded from analysis, the median number of patients tested for chlamydia increased to 0.92 per index case (1.00 in Greater London, 0.89 in all other clinics).

FIGURE 5.

Control chart for partners tested for chlamydia per index case. Combined outcome was defined as outcome verified by a health-care worker or, if data were missing, as reported by the patient. All cases were included and those with missing data were coded as having zero partners tested.

The other measures of partner notification outcome were the number of partners per index case with a positive chlamydia test (Figure 6, median 0.37 assuming missing values as 0 partners, and 0.75 excluding missing values) and the number of partners per index case treated for chlamydia (Figure 7) (median 0.60 and 0.95).

FIGURE 6.

Control chart for partners tested positive for chlamydia per index case. Combined outcome was defined as outcome verified by a health-care worker or, if data were missing, as reported by the patient. All cases were included and those with missing data were coded as having zero partners tested.

FIGURE 7.

Control chart for partners treated for chlamydia per index case. Combined outcome was defined as outcome verified by a health-care worker or, if data were missing, as reported by the patient. All cases were included and those with missing data were coded as having zero partners tested.

Discussion

This study showed marked variation between genitourinary medicine clinics in outcomes of partner notification for chlamydia, most resulting from common causes. Men who have sex with men were less likely than heterosexual men and symptomatic patients less likely than asymptomatic patients to have had at least one partner tested for chlamydia. In clinics diagnosing greater numbers of chlamydia cases the odds of recording at least one partner tested for chlamydia were higher than in smaller clinics. Findings were similar with outcomes of the number of partners with a positive chlamydia test and number of partners treated for chlamydia.

The main strength of this study was that it included information from about 5000 patients with chlamydia in genitourinary medicine clinics across the UK. The main limitation of the study was related to the high level of missing values for numbers of contacts, which resulted in a twofold difference in calculated values for the outcome. Differences between index cases with and without information about the outcomes of partner notification mean that the observed outcomes are likely to be biased but it is not possible to state the direction or degree. A further limitation was that the method of partner notification for each index case was not recorded. We assume that the majority of partners were notified following patient referral by the index case him or herself.

In this study, we were able to investigate individual- and clinic-level factors influencing common causes69 of the variation in partner notification outcomes. Men who have sex with men with gonorrhoea have previously been observed to have fewer partners tested or treated, despite reporting higher numbers of partners than heterosexual men. 72 Symptomatic patients also had poorer partner notification outcomes than asymptomatic patients. This might be related to the differing look-back periods of 1 month for symptomatic patients and 6 months for asymptomatic patients (as recommended by the Society for Sexual Health Advisers73). Additionally, symptomatic and asymptomatic cases may differ in their relationship to, and ability to contact, sexual partners, although data were not collected about these factors. The observed association between numbers of chlamydia cases diagnosed and partner notification outcomes was difficult to interpret. This finding might reflect biases for which we could not control in the analysis, for example if patients in larger clinics reported more partners, or as a result of missing data, which was more common in large clinics. There was a strong influence of missing information about partner notification outcomes on the denominator used to calculate the mean for each clinic, which then influenced the estimates of the overall median: 0.47 partners tested for chlamydia per index case when those with missing data were coded as having no partners tested and 0.92 when missing values were excluded.

There was wide variation in partner notification outcomes across clinics, from 0 to 1.5 partners per index case tested for chlamydia. The median number of partners per index case reported to have been treated in this audit was similar to the median of the values observed in randomised controlled trials of a variety of partner notification technologies (see Table 7). The extent to which unintended variation in clinical outcomes can and should be eliminated has been debated. 69,74,75 If clinics took further measures to improve their performance this would improve both the consistency and the quality of clinical care. The data provided by this audit can be used in mathematical modelling studies as baseline values representing current practice for patient referral. In practice, the completeness of recording needs to improve so that levels of partner notification outcomes can be measured more accurately in future. Further research is needed to identify auditable measures that are associated with successful partner notification that prevents repeated chlamydial infection in index cases.

Introduction

The previous two sections present clinical outcome measures of partner notification, focusing on the incidence of reinfection in index cases and the number of sexual partners treated per index case. These outcomes are not usually reported according to the type of sexual partnership between an index case and their partner(s). Partnership type, however, is likely to determine whether index cases operate as ‘spread’ networks or ‘dead–end’ networks for STI transmission. 5 The impact of partner notification can therefore differ according to which partner types are sought. Dynamic mathematical models (see Chapter 3) are used to examine the potential epidemiological impact of partner notification at the population level, but they are too complicated to be used by individual clinical services planning or assessing the impact of partner notification activity on local transmission patterns. Here, we use a simple algorithm that clinicians and public health teams can use, without the need for onerous data collection or computational methods. 9

Objective

To estimate the likely short-term impact of partner notification activity on preventing STI transmission according to partner type, using data that are routinely collected in genitourinary medicine clinics.

Methods

The authors developed an algorithm to estimate the numbers of cases of chlamydia transmitted to the sexual partners of index cases and the number of secondary chlamydia transmissions from the partners of the index cases. They then estimated the impact of partner notification on preventing chlamydia transmission. The development of the methods used has been described in detail. 9 Here we applied the methods using audit data collected for this project about patients diagnosed with chlamydia study at one genitourinary medicine clinic in England, and estimates of C. trachomatis transmissibility that were developed for this study (see Modelling the transmission dynamics of C. trachomatis).

Results

Estimating numbers of primary transmissions

The 203 index cases (112 men and 91 women) reported a total of 346 partners in the past year (Figure 8). Using the per-partnership probabilities of transmission described above, we estimate that a total of 190 primary chlamydia transmissions occurred between the index cases and their 346 partners.

FIGURE 8.

Frequency distribution of partners over the past year, by partnership type, sex and age group of index case. Numbers in parentheses correspond to the number of index cases in each group.

Discussion

This paper has described an evidence-based algorithm that clinicians and public health teams can use to estimate the potential impact of their service’s partner notification effort on preventing STI transmission. Additionally, we have shown how the potential impact of partner notification can be quantified for different patient groups by using two new epidemiological measures: the AROT and the NNTIT.

The strength of this study is that routinely collected data can be used to adapt the algorithm to different settings and populations. In contrast to the published paper,9 we used data on actual partner numbers as reported in health adviser interviews, rather than estimates from Natsal-2. However, index cases reported their partners by type, which was subjective, with women more likely than men, for example, to regard a partnership as regular rather than casual. 81 There are limitations inherent to the static deterministic modelling approach used here. The algorithm makes a number of assumptions about the transmission of chlamydia to secondary partners. Furthermore, we assumed that these potentially infectious encounters were all susceptible to chlamydia (i.e. not already infected or immune), which is highly unlikely in contacts who are closely linked to a case. It is not possible to infer whether or not the estimates of onward transmission derived here are optimistic, as counting further generations of cases would increase the estimate, whereas pre-existing infection, chance effects, and network structure would tend to decrease the effect. The estimates of onward transmission given here should, therefore, be taken as broadly indicative rather than precise. The transmission dynamic models used in Chapter 3 overcome these limitations because they have partnerships explicitly modelled and take into account infection status.

There are additional limitations to the study. We treat age crudely, stratifying into two age groups, though using 25 years old as a threshold does correspond to age groups currently used for UK surveillance. There were insufficient data in Natsal-2 to allow stratification by ethnicity and sexual orientation, which are also associated with both partner numbers and the probability of STI acquisition, though other data sources could also be harnessed. Furthermore, we assumed assortative mixing by behaviour and partnership type, owing to the difficulty of obtaining data on partners’ current behaviours and partnerships. This might be a reasonable assumption given the extent of assortative mixing in the population. 82 Although there is uncertainty and variability surrounding the parameters used in our algorithm, sensitivity analyses undertaken to explore the vulnerability of the algorithm to uncertainty were reassuring.

As provider referral (vs. patient-led partner notification) is required more often with casual and ex-partners than with regular partners (Gill Bell, Sheffield Teaching Hospitals NHS Foundation Trust, 2012, personal communication), realising the potential public health benefits suggested by the NNTIT has cost implications for services. However, as casual and ex-partners are likely to have greater numbers of partners themselves, the potential for preventing onward transmission is greater, yielding greater public health benefit.

Services should collect data from their index cases on the number and type(s) of partners, as well as the partner notification method(s) required (patient vs. provider referral). These data, when combined with cost data on partner notification, are key for demonstrating the cost–benefit of provider referral in preventing onward transmission in a local population.

Our analyses show that future debate and research about partner notification provision need to develop a partnership-orientated focus regarding which index cases should be offered more intensive support for public health benefit. The existing emphasis on different strategies for partner notification (e.g. patient, provider) has a service-orientated focus regarding the types of partner notification services offered. This now needs to be explicitly linked to strategic decisions about what index cases should be offered, according to their different kinds of partnership history. Although clinicians recognise that a ‘one size’ approach to partner notification does not fit all types of sexual partner, outcome measures must also reflect the public health importance of partnership type. Patient-centric measures conceal huge variations in transmission potential by partner type. Clinical services should routinely adopt partner-centric outcome data, and audit their partner notification data by partner type. This will allow them to explore the implications of different approaches to targeting partner notification activity, and assess the public health outcomes of the service.

Our partner notification algorithm has a role to play in providing epidemiological evidence aimed at assessing and justifying the public health value of more expensive and challenging partner notification activity with casual and former partners, which is provided to a variable extent. 83 The algorithm can be adapted to other STIs, such as gonorrhoea, and other infections where partner notification is needed, such as hepatitis B and tuberculosis. Without a better understanding of how to harness the relative transmission prevention potential of different partnership types, evidence-based partner notification practice cannot progress.

Introduction

We need to understand the transmission dynamics of STIs if we are to make accurate quantitative, as well as qualitative, predictions about the impact of partner notification and other preventative interventions. Sexual partnership dynamics and the values for infection parameters are key determinants that affect the spread of STIs. In this chapter, we investigate the effect of different assumptions about disease-specific parameters for C. trachomatis on the predicted impact of interventions using a basic transmission model. We then describe the derivation of improved estimates of the infectious duration and the transmissibility of chlamydia. Finally, we present the results of a detailed comparison of the three individual-based models, which allows us to draw more robust conclusions about the effect of screening and partner notification interventions on chlamydia transmission. We end with a brief summary of the insights that we have gained into chlamydia transmission models, and a rationale for the model choice in subsequent chapters.

Infectious duration of chlamydia

Different models of chlamydia transmission use widely different disease-specific parameters. We devised a deterministic chlamydia transmission model with a SEIRS (susceptible-exposed-infected-recovered-susceptible) structure. For simplicity, we assumed a closed population of young adults and a homogeneous population where both sexes become infected and pass through the infected stages at equal rates. The full details of the model can be found in Althaus et al. 46

To understand what parameters are most important for predicting the impact of an intervention we performed a univariate sensitivity analysis of disease-specific parameters, namely the duration of the asymptomatic and symptomatic periods, the duration of temporary immunity and the fraction of infections that become symptomatic. The pre-intervention prevalence of chlamydia in the population was assumed to be 5%. This corresponds roughly to the prevalence observed in sexually active young adults. 77

We investigated the impact of a screening intervention rather than partner notification because the simplicity of the model does not allow identification and tracking of current and previous sex partners of infected index cases. We then calculated the expected prevalence after the introduction of two different screening scenarios: (1) an organised screening programme with a screening rate of 0.25 per year implemented for 10 years and (2) opportunistic screening at a rate of 0.05 per year, in which the new steady-state prevalence is shown after long-term implementation.

Uncertainties in the duration of asymptomatic period and of temporary immunity resulted in large differences in the predicted impact of a screening programme (Figure 9a and b). In contrast, varying the fraction of infections becoming symptomatic and the duration of the symptomatic period within the range of previously used parameter estimates caused little change in the predicted effect of interventions.

FIGURE 9.

Prevalence of chlamydia after the introduction of a screening programme. Dotted line, baseline prevalence in the absence of a screening programme; dashed line, long-term prevalence if the population receives screening at a rate of 0.05 per year; solid line, prevalence after screening the population at a rate of 0.25 per year for 10 years; grey area, parameter range; black dots, baseline scenario. (a) Chlamydia prevalence as a function of the duration of the asymptomatic period. Most estimates on the duration of the asymptomatic period are within 200–400 days; and (b) chlamydia prevalence as a function of the duration of temporary immunity. Reprinted from Epidemics, vol. 2, Althaus CL, Heijne JCM, Roellin A, Low N, Transmission dynamics of Chlamydia trachomatis affect the impact of screening programme, pp. 123–31, 2010, with permission from Elsevier. 46

The long-term outcome of a screening programme appeared to be most sensitive to the duration of the asymptomatic period. Figure 9a (grey area) shows the range of values for the asymptomatic period that have been used in previous mathematical models of chlamydia transmission (180–420 days). The long-term impact of screening at a low rate (dotted line) is much more pronounced if the asymptomatic period is at the upper bound of the range.

The duration of temporary immunity also affected the predicted impact of screening. Increasing the duration of temporary immunity substantially decreased the impact of screening (see Figure 9b). Here, screening and treating asymptomatically infected people prevents the development of temporary immunity and renders them immediately susceptible. This somewhat counterbalances the otherwise strong impact of screening. The grey area, which covers the range of values used in previous modelling studies, shows the uncertainty about the existence and duration of immunity after a chlamydia infection.

To obtain a robust estimate of the duration of asymptomatic chlamydia infection in women, we fitted a mathematical model to data from a study that followed a large number of asymptomatic chlamydia-infected women and showed that the infection can persist for several years (Figure 10). 85 The model describes the clearance of chlamydia infections in all women who were infected at the beginning of the study. We also took into account the possibility that women could have cleared their infection naturally and become reinfected from their (presumed) untreated partners. Full details of the model and the parameter estimation have been published. 46

FIGURE 10.

Persistence of chlamydia in asymptomatically infected women. Data from Molano et al. 85 and reprinted from Epidemics, vol. 2, Althaus CL, Heijne JCM, Roellin A, Low N, Transmission dynamics of Chlamydia trachomatis affect the impact of screening programme, pp. 123–31, 2010, with permission from Elsevier. 46

The estimated reinfection rate is low (0.01 per year, 95% CI – 0.01 to 0.03 per year), which indicates that the data are mainly described by natural clearance. With an estimated clearance rate of 0.84 per year (95% CI 0.82 to 0.87 per year), we obtain a mean duration of the asymptomatic period of 433 days (95% CI 420 to 447 days).

Transmissibility of chlamydia

The transmissibility of C. trachomatis through sexual intercourse cannot be observed in empirical studies. Estimates of the transmissibility of chlamydia from epidemiological studies have usually been based on data from couples with and without chlamydia infection and the proportions of concordant and discordant pairs. 6,7,86 Katz86 proposed an original approach for analysing such data using information about heterosexual couples attending a STI clinic in Indianapolis, IN, USA. The expected numbers of concordant and discordant couples before transmission takes place can be calculated if it is assumed that all couples in the population with at least one infected individual have the same probability of observation, and that sexual partnership formation is independent of infection status. After sexual partnerships have formed, transmission can happen in discordant partnerships, resulting in a higher proportion of couples in which both partners are positive.

Katz86 estimated the probabilities that transmission occurred within a couple at 0.395 from men to women and 0.323 from women to men. But there are two major problems with this approach. First, the estimated transmission probabilities do not represent the per-partnership transmission probability because infection status is observed during the partnership and not at the end. The probabilities estimated by Katz are often called per-partnership transmission probabilities, which imply that the partnership has been observed until it has ended. Second, they do not take into account the natural history of chlamydia infection, where spontaneous clearance and reinfection within sexual partnerships can occur. 87 These complexities need to be considered because different assumptions about infectious duration46 and reinfection in sexual partnerships88 can affect the prevalence of chlamydia.

To obtain new estimates of the transmissibility of chlamydia we used data from a cross-sectional partnership study that has frequently been cited as the source of estimates for chlamydia transmissibility. 7 We described the transmission of chlamydia using the pair model framework, which has been used for several STIs. 88–91 The full details of the model and the parameter estimation procedure have been published. 47 In brief, we used maximum likelihood estimation92 to fit the model to the data from the study by Quinn et al. 7 The study contains information about chlamydia infection status and sexual activity in 494 heterosexual couples. The study reported 53 concordant chlamydia positive, 48 discordant and 393 concordant negative couples. We generated 1000 parameter combinations to account for uncertainties in the duration of sexual partnerships and infections. Sexual behaviour parameters were taken from the same study. We obtained model estimates of the per partnership transmission probability for different values of the number of partners during the last 6 months (Figure 11a).

Higher number of partners resulted in lower estimates of the per partnership transmission probability. However, partner numbers of three or more during the last 6 months resulted in poor fits to the data. We therefore consider two partners during the last 6 months as our baseline scenario, for which the median of the estimated per-partnership transmission probability is 55.5% [interquartile range (IQR) 49.2–62.5%]. The estimates of the per-sex act transmission probability seemed to be less affected by the assumed number of partners in the last 6 months (Figure 11b). Most values were around 10% with the median of the baseline scenario at 9.5% (IQR 6.0–16.7%).

FIGURE 11.

Estimated chlamydia transmission probabilities for different values of the number of partners in the last 6 months. (a) Per-partnership transmission probability of chlamydia; and (b) per-sex act transmission probability of chlamydia. Each box plot represents estimates from 1000 different parameter combinations. The baseline scenario, where it is assumed that individuals in a partnership at steady-state have on average of two heterosexual partners during the previous 6 months, is in grey. It is assumed that individuals have one episode of heterosexual intercourse every 5 days. Reprinted with permission from Althaus CL, Heijne JC, Low N, Towards more robust estimates of the transmissibility of Chlamydia trachomatis. Sex Transm Dis, vol. 39, issue 5, pp. 402–4, 2012. 47

Our estimated range of chlamydia transmission probabilities in heterosexual partnerships is higher than the baseline values reported by Katz. 86 This is expected because we report the probability of transmission taking place by the end of a partnership and we assume that chlamydia can be cleared spontaneously followed by reinfection. Our estimate is, however, lower than the percentages of concordantly infected couples (70% of female partners of infected men and 68% of male partners of infected women). These percentages are often interpreted as the per-partnership transmission probabilities,84,93 but they are not because the direction of transmission cannot be reliably determined. 87 This discrepancy illustrates the importance of taking the natural history of chlamydia infection and the dynamics of sexual partnership formation into account when estimating transmissibility from data of chlamydia positivity in couples. The refined estimates of the per-sex act transmission probability for chlamydia are consistent with those obtained or used in other modelling studies. 35,94,95 In two of the three individual-based models of chlamydia transmission that we had compared, the per-sex act transmission probabilities were also within the same range as our estimate, but the probability in the third model (3.75%) was lower.

Comparison between three individual-based models

We evaluated further the three individual-based models of chlamydia transmission that showed conflicting results about the impact of the same screening intervention. 41 The names of the models are the same as those used in the first comparison: the ClaSS (Chlamydia Screening Studies project) model and Turner model were both developed to examine the effects of different models of chlamydia screening in the UK; the Kretzschmar model was developed in the Netherlands. The comparison has been published in full. 48

First, we compared the sexual partnership dynamics of the models to population-based data from Natsal-2 about the durations of sexual partnerships, the length of gaps and overlaps between partnership and the numbers of partners in the last year. We showed that although all models capture some aspects of the sexual partnership dynamics reasonably, they fail in others. Overall, the Kretzschmar model performed best.

We then investigated the spread of chlamydia in the simulated populations with Natsal-2 (Figure 12). In previous studies, the three models used different assumptions about the duration of infection and the proportion of symptomatic and asymptomatic cases. 41 To better compare their results, we used a harmonised set of chlamydia infection parameters, based on a thorough evaluation of the parameter ranges in the literature and our new estimate of the duration of the asymptomatic period.

FIGURE 12.

Asymmetrical distributions of chlamydia infections among individuals according to level of sexual activity. (a) Natsal-2;8 (b) Kretzschmar model;38 (c) Turner model;40 and (d) ClaSS model. 39 The width of each bar represents the proportion of people who have had a given number of sexual partners within the last year (the legend shows the distribution of sexual partner numbers). The height of the bar indicates the prevalence of chlamydia in each group. Note that the prevalence in the group with highest activity (> 5 partners) is 20% in the Kretzschmar model and 34% in the ClaSS model. Overall prevalence for women and men aged 18–44 years is given by the dashed line and is approximately the same in all graphs (1.7%). Reproduced with permission from Althaus CL, Turner KM, Schmid BV, Heijne JC, Kretzschmar M, Low N, Transmission of Chlamydia trachomatis through sexual partnerships: a comparison between three individual-based models and empirical data, J R Soc Interface 2012; 9:136–46. 48

The per-sex act transmission probabilities required to reproduce the estimated prevalence from Natsal-2 are lower than the estimates we obtained (see Transmissibility of chlamydia). This might be the result of unrealistic assumptions about the sexual partnership dynamics in the three models. However, it could also derive from our assumption of a relatively high frequency of sex acts during the first 2 weeks of a sexual partnership, which would balance the lower transmission probabilities per sex act.

We compared the model predictions with Natsal-2 data according to the distribution of chlamydia infections according to numbers of sexual partners and the fraction of infections in individuals with frequent partner change rates. As expected, the prevalence of chlamydia increases with an increasing number of heterosexual partners within the last year (see Figure 12). Again, the Kretzschmar model captures the overall distribution well and bears a striking resemblance to the proportion of people in each ‘risk category’ (shown in inset legend for each panel). The prevalence in the group of individuals with five or more partners within the last year is, however, overestimated. The Turner model captures the overall distribution of chlamydia infections in 18- to 44-year-olds but the figure excludes 16- to 17-year-olds (because they were not included in Natsal-2) in whom chlamydia prevalence in the model was very high. Furthermore, the mode results in an unrealistic outcome where no individual remains without a partner throughout the 1-year period. In contrast to the other two models, chlamydia infections are too heavily concentrated among ‘high-risk’ individuals in the ClaSS model. Furthermore, it becomes apparent that sexual partnership dynamics are not captured well because too many individuals remain single throughout the 1-year period.

Finally, we implemented a standardised screening and partner notification intervention and compared the predicted impact after 5 and 10 years of intervention (Figure 13).

FIGURE 13.

Impact of a standardised screening programme on the prevalence of chlamydia. The proportion of eligible people receiving screening at least once a year is 20%, and 40% of the current or most recent partners of an index case are successfully notified and treated. Means of 100 simulations runs are shown together with the SD. Reproduced with permission from Althaus CL, Turner KM, Schmid BV, Heijne JC, Kretzschmar M, Low N, Transmission of Chlamydia trachomatis through sexual partnerships: a comparison between three individual-based models and empirical data, J R Soc Interface 2012;9:136–46. 48

The harmonisation of disease-specific parameters resulted in more similar results for the Kretzschmar and the ClaSS models. However, the Turner model still predicted a much stronger impact of the same intervention. Part of this discrepancy can be explained by an overestimation of partner change rates among young adults in the Turner model. The unrealistically high rates cause a large proportion of chlamydia transmission to occur in exactly those individuals who become eligible for screening, rendering the intervention more powerful.

Discussion

We re-estimated the average duration of asymptomatic chlamydia infection in women at 433 (95% CI 420 to 447) days. We also obtained new estimates of chlamydia transmissibility and found that the transmission probability per episode of heterosexual intercourse is 9.5% (IQR 6.0–16.7%). The transmission probability per heterosexual partnership was estimated at 55.5% (IQR 49.2–62.5%). We applied this parameter set to three individual-based models of chlamydia transmission and obtained similar predictions of the impact of a screening and partner notification intervention in two of the models. Part of the discrepancies between the models was explained by their use of different assumptions about the infectious duration of chlamydia.

To fully evaluate the role of immunity on the impact of chlamydia prevention, we need further insights about the possibility of temporary immunity to C. trachomatis in humans and the timing of its development. The development of immunity could interfere with the effect of screening on reducing the prevalence of chlamydia. Whether or not natural clearance of asymptomatic infection is followed by a period of temporary (or partial) immunity is still open to debate. 96 We found that long duration of temporary immunity can drastically diminish the effect of screening. If temporary immunity develops only in asymptomatic individuals who clear the infection naturally, screening and treatment might directly interfere with establishing immunity, causing a diminished effect of screening. 36,96,97

The type and structure of models that describe the spread of STIs should depend on the question that one wants to address. 98 The duration and transmissibility of chlamydia infection were investigated with relatively simple deterministic, population-based models. These models do not allow the identification and tracking of current and previous sex partners of infected index cases, which is needed for detailed investigation of the effects of partner notification. The three models that we investigated differed in their ability to realistically describe the sexual partnership dynamics. Overall, the Kretzschmar model performed best in terms of describing the sexual partnership dynamics and the distribution of chlamydia within the population. The Turner model generated a good description of the distribution of chlamydia in the age range investigated, but probably overestimated the effect of chlamydia screening due to an unrealistically high prevalence in the youngest age class. The ClaSS model predictions of the impact of chlamydia screening were similar to those of the Kretzschmar model after harmonisation of parameters, but the distribution of chlamydia was too highly concentrated among those with a high number of sexual partners.

Based on these studies, we decided not use either the Turner or the ClaSS models to study the effects of partner notification interventions. We developed new individual-based models, which were simple enough to allow interpretation of the findings, but which possessed the complexity necessary to construct the sexual network and track individual partnerships.

Introduction

Partner notification is known to be an efficient method of case finding. There is, however, no consensus about the most appropriate or efficient window of time for past partner notification (‘look-back period’). The UK National Guideline for the Management of Genital Tract Infection with Chlamydia trachomatis recommends notifying partners of an asymptomatic index case within a period of 6 months. 10 The same is standard in Sweden, but recommendations from the Swedish National Board of Health and Welfare are changing, based on a recent study that found that extending partner notification periods could improve the identification of new chlamydia cases. 99 This is also supported by a study from the USA, where it was observed that a partner notification period of 6 months or more would help to identify more infected cases. 100 But the Sexually Transmitted Diseases Treatment Guidelines from the US Centers for Disease Control and Prevention recommend notifying partners with whom the index case has had sexual contact within the previous 60 days, or the most recent partner if no sexual contact occurred in this period. 101

Objectives

To investigate the impact of two partner notification strategies – notifying partners in order of the most recent date of sexual intercourse before the end of a partnership, or notifying all partners during a specified look-back period – on outcomes at the level of individuals and the population.

Methods

Sexual network

Figure 14 shows the sexual partnership dynamics of the model where chains of contacts can occur at cross-section. Over the course of 1 year, the sexual partnership network shows closely connected groups or bigger circular structures.

FIGURE 14.

Sexual partnership networks from the individual-based model. Different sexes are indicated by filled and empty circles. (a) At cross-section, individuals can be single, form a sexual partnership or have two concurrent sexual partnerships; and (b) large connected components emerge during a period of 1 year. For illustrative purposes, the population size was limited to 100, resulting in higher connected networks compared with larger population sizes.

Partner notification strategies

Using our individual-based modelling framework, we can follow an individual’s history of current and previous partners. Figure 15 shows the complex partnership dynamics in which a new partnership can replace a previous one, or where short episodes of concurrency can occur.

FIGURE 15.

Partnership histories and partner notification strategies. The scheme shows an example of an individual’s history of three sexual partnerships. The partnerships can either be separated by a period of being single (partnership 1 and 3) or be concurrent (partnership 1 and 2). One strategy is to notify partners of an index case in order of their recency, e.g. the first, second or third most recent partner(s). Another strategy is to notify all partners from a certain time period, e.g. the partners from partnership 1 and 2 only. Reprinted with permission from Althaus CL, Heijne JCM, Herzog SA, Roellin A, Low N, Individual and population level effects of partner notification for Chlamydia trachomatis, PLOS ONE 2012; 7:e51438. 49

We explore two strategies in which partners of an index case can be notified. One strategy is to notify the partners in order of their recency (by the time since their partnership ended). The other strategy is to notify all partners within a certain look-back period. For each strategy, each notified partner is tested and successfully treated with a probability of 50% (approximated from results; see Partner notification outcomes for chlamydia in UK genitourinary medicine clinics). 45 In practice, a partner who tests positive could become a new index case. We did not include this feature because it made the model more complicated but did not affect the results.

Results

Individual-level effect of partner notification

At the level of individuals, partner notification identifies new index cases. It is therefore important to know how many of an index case’s partners are infected. A Swedish study99 found that the proportion of chlamydia-positive partners decreases as time since last sexual intercourse increases. In the simulations, everyone who is chlamydia positive at cross-section is defined as an index case. This corresponds to infected individuals who would be detected through random screening or after developing symptoms, assuming that symptoms occur uniformly throughout the course of infection. We can then go through all current and previous partners of an index case and ‘test’ whether or not they are infected. When we used the same time periods as in the Carré et al. study,99 chlamydia positivity in the model was similar to the data (Figure 16).

FIGURE 16.

Simulated and empirical data of chlamydia positivity in partners of index cases. Black crosses correspond to the proportion (with 95% CI) of positive partners out of those with a positive test. 99 The black circles represent simulated data from the model. In the simulations, the steady-state prevalence of chlamydia is 3%. Means of 100 simulation runs are shown. Standard errors are small and omitted for better visibility. Reprinted with permission from Althaus CL, Heijne JCM, Herzog SA, Roellin A, Low N, Individual and population level effects of partner notification for Chlamydia trachomatis, PLOS ONE 2012; 7:e51438. 49

The proportions of infected cases were examined in more detail in the individual-based modelling framework. First, we investigated in order of recency the proportion of infected partners (Figure 17a). The model gave consistent results: 67.5% of the most recent partners of an index case were infected with chlamydia. The less recent a partnership or contact, the lower the probability that the partner is infected. The proportion of infected individuals, up to the third most recent partner, was still substantially higher than the population prevalence.

We could also group the partners of an index case in a look-back period since the partnership has ended in greater detail than that shown by Carré et al. 99 (Figure 17b). We found that as far back as 18 months, a substantial proportion of partners (> 10%) were infected with chlamydia. This suggests that extending partner notification periods beyond 1 year yields more new index cases for individual case management than would be found through random screening.

FIGURE 17.

Proportion of chlamydia-positive partners of index cases. (a) The proportion of infected partners in order of their recency; and (b) the proportion of infected partners grouped into different look-back periods in order of their break-up date. Steady-state chlamydia prevalence is 3% (dashed line). For each strategy, means of 100 simulation runs are shown. Standard errors are small and omitted for better visibility. Reprinted with permission from Althaus CL, Heijne JCM, Herzog SA, Roellin A, Low N, Individual and population level effects of partner notification for Chlamydia trachomatis, PLOS ONE 2012; 7:e51438. 49

Population-level effect of partner notification

At the population level, partner notification can prevent onward transmission of chlamydia and reduce the overall prevalence of the infection. Here, we investigated the effects of the different partner notification strategies if they were implemented as part of a population-wide screening programme. After the simulations approached the steady-state prevalence of 3%, we introduced random screening of the whole population of young adults. Every woman and man received screening at a rate of 0.1 per year, that is, every 10 years on average. If partner notification is performed, each notified partner is tested and successfully treated with a probability of 50% (see Chapter 2, Clinical effectiveness of partner notification technologies: systematic review). 45

After 5 years of screening, assuming that there is no partner notification, the prevalence of chlamydia was reduced to about 70% (Figure 18). It can be clearly seen that the strongest effect of partner notification stems from notifying only the current partner. This suggests that notification of current partners, or the most recent partners if the index case is single, is sufficient to achieve most of the additional reduction in prevalence at the population level.

FIGURE 18.

Population level effect of partner notification. Graphs show the reduction in the prevalence of chlamydia after screening for 5 years at a rate of 0.1 or 0.25 per year. (a), (c) and (e) show prevalence for different numbers of notified partners; (b), (d) and (f) show prevalence for different partner notification periods. The probability, p, that each notified partner will be tested and successfully treated is set to 10% (a and b), 50% (c and d) or 90% (e and f). For each strategy, means of 100 simulation runs are shown. Standard errors are small and omitted for better visibility.

Discussion

Our model simulations show that, while extending partner notification periods beyond 1 year helps to find new index cases, notifying the current or most recent partner has the greatest effect on reducing transmission in a general heterosexual population of young adults.

We made a number of simplifying assumptions in our model. First, we treat the population of 16- to 25-year-olds as a closed population that mixes homogeneously. This assumption probably resulted in realistic chlamydia transmission dynamics in our study because it was restricted to the age group that drives chlamydia transmission in the wider population. We have previously shown that this kind of model results in a realistic distribution of chlamydia infection according to numbers of recent sexual partners,46 and provides conclusions in line with more detailed models that include age and risk stratification. 35,48 However, differences in sexual activity between women and men, and potentially higher concurrent partnerships in men compared with women, are additional complexities that could be considered in future studies.

Our simulation study examined a general strategy for partner notification, which is similar to traditional technologies where the partner attends a health service setting and receives both testing and treatment. Our findings suggest that the choice of partner notification strategy for chlamydia in a general heterosexual population of young adults depends on the public health context in which it is applied. At the individual level, our results suggest that tracing as many as three partners from the preceding 18 months can be helpful in finding new index cases. This is in line with recent recommendations for partner notification periods in Sweden, based on the findings of Carré et al. 99 At the population level, notification of current partners of an index case should be a priority. But if previous partners are chlamydia positive, they are likely to have been infected for a long time and thus notifying them will do little to limit onward transmission in a general heterosexual population with limited levels of concurrent partnerships. In all likelihood, they will have already transmitted the infection to others before clearing it spontaneously or becoming symptomatic and seeking care. If screening is targeted at high-risk individuals, notifying previous partners of index cases will likely have a stronger effect on limiting onward transmission because of their higher partner change rates (see Estimating the likely public health impact of partner notification for a clinical service: an evidence-based algorithm). 9

Few modelling studies have examined the effects of partner notification for chlamydia. Our findings are consistent with those of Kretzschmar et al. ,84,104 who also considered a dynamic sexual partnership network but restricted partner notification to current partners only. In contrast, Armbruster and Brandeau105 found that increasing contact tracing capacity resulted in a substantial reduction of chlamydia prevalence, although with diminishing returns. They investigated the effects of partner notification for index cases who seek treatment for symptoms and assumed a static sexual network with a relatively high prevalence of chlamydia. In our study using a dynamic sexual network, we found that previous partners of an index case do not contribute to reinfection, which might explain our finding that notifying previous partners had little additional effect on reducing chlamydia transmission.

In future modelling studies, different strategies for screening and partner notification in groups with frequent partner change rates could be investigated in models that incorporate heterogeneity in sexual behaviour. The effects of different partner strategies for other bacterial STIs should also be examined. In the following section, we investigate testing and treatment strategies for both chlamydia and gonorrhoea. The impact of partner notification on syphilis would require a different model structure and should be examined in a population of men who have sex with men, or heterosexuals with higher average levels of sexual behaviour. In this study, we have shown that notifying three or more partners from the preceding 18 months is expected to yield substantial numbers of new chlamydia cases. In contrast, the successful treatment of current partners is most important for preventing reinfection of index cases and reducing further transmission of chlamydia at the population level. In the next section we therefore focus on the effects of partner notification for the current or most recent sexual partner.

Introduction

The second most commonly diagnosed bacterial STI after chlamydia is gonorrhoea (see Table 3). In the previous chapters, we investigated the effects of traditional technologies for partner notification for index cases with chlamydia only. We have assumed that notified partners will attend a health service setting and be tested for other STIs. New partner notification technologies, such as APT, which allow partners to collect treatment without attending a health service setting, might leave partners with undiagnosed STIs. It is therefore important to study the potential effects of cotesting and cotreatment for chlamydia and gonorrhoea.

An important observation for studying combined interventions against both chlamydia and gonorrhoea is the high rate of chlamydia infection in those who are infected with gonorrhoea. This is a consistent finding in studies of different populations and with different study designs. For example, Nsuami et al. 27 found that 41.4% of female and 46.7% of male students tested positive for both gonorrhoea and chlamydia in an urban school district in the USA where STI testing was offered systematically every year. In the US National Health and Nutrition Examination Survey, a nationally representative population-based study, the chlamydia co-infection rate was 45.7% in 14- to 39-year-old civilians with gonorrhoea infections. 26 Similar levels of co-infection are found in genitourinary medicine clinic attenders. 25 These rates are much higher than the population prevalence of chlamydia and even exceed the positivity rates of chlamydia in high-risk populations.

It is important to investigate potential mechanisms for chlamydia and gonorrhoea co-infection on the transmission dynamics in mathematical models. One hypothesis is that biological interactions between the two organisms account for the high rates of co-infection. If such interactions exist, then understanding them would enable us to improve guidelines for testing and treating a population for chlamydia and gonorrhoea. Classical models, where chlamydia and gonorrhoea transmission does not depend on the presence of another organism, generally fail to account for high co-infection rates. 106

Objectives

(1) To develop a dynamic transmission model of chlamydia and gonorrhoea infection in a heterosexual population and (2) to use the model to investigate the effects of APT and traditional partner notification technologies on STI prevalence, the frequency of gonorrhoea outbreaks and STI reinfection rates.

Methods

We examined epidemiological data to inform the development of a mathematical model in which the presence of one organism can increase the susceptibility and the transmissibility of the other. The deterministic, population-based model was then implemented in an individual-based manner using Rstisim and we studied basic aspects of the effects of different testing and treatment strategies in separate models that either assumed that chlamydia and gonorrhoea are transmitted independently or included an interaction between the organisms.

Results

Co-infection with chlamydia and gonorrhoea

Figure 19 shows the predicted percentages of the model population infected with chlamydia and gonorrhoea. If chlamydia and gonorrhoea transmit independently (0% increase in susceptibility and transmissibility), 25.2% of those infected with gonorrhoea are also infected with chlamydia and the prevalence of gonorrhoea in those infected with chlamydia is 4.2%. These percentages are lower than those observed in the empirical studies. 26,27 The rates of co-infection increase, particularly for co-infection with chlamydia in those who are infected with gonorrhoea, when we increase the susceptibility and transmissibility in the model. If the risk of transmitting and acquiring one infection when the other infection is also present is increased by 50%, the predicted co-infection rate increases to 36.8% and 6.2%, respectively, and are more consistent with the observed data. 26,27

FIGURE 19.

Rates of co-infection with chlamydia and gonorrhoea. Increasing the susceptibility and transmissibility of one organism if the other organism is present results in an increase in co-infection rates, particularly the rate of chlamydia positivity in those who are infected with gonorrhoea.

We consider two scenarios for investigating the effects of cotreatment for chlamydia and gonorrhoea: a scenario with no interaction between the two infections and a scenario in which the presence of the other organism enhances the transmissibility and the susceptibility of an infection by 50%.

Screening and partner notification for chlamydia and gonorrhoea

We investigated the effects of testing and treating index patients and their partners for chlamydia and gonorrhoea (Table 14). Partner notification takes place as an integral part of a screening strategy, which detects index patients. The combined intervention results in a reduction in the prevalence of chlamydia and gonorrhoea. The new steady-state prevalence is usually reached about 5 years after starting the intervention.

TABLE 14 - Testing and treatment strategies for index cases and their most recent partners

The percentages represent the probabilities of successful treatment. Note that this is a combination of the treatment in index cases and the possibility of reinfection by their partners.

Scenario	Treated successfully (%)
	Index patient		Most recent partner
	Chlamydia	Gonorrhoea	Chlamydia	Gonorrhoea
1: Current practice for chlamydia	70	0	30	0
2: APT for chlamydia	90	0	60	0
3: Routine testing for chlamydia and gonorrhoea	70	70	30	30
4: APT for chlamydia and gonorrhoea	90	90	60	60

The probability of successful treatment is estimated as the product of antibiotic treatment efficacy and the probability that the patient becomes reinfected by an untreated partner. APT is assumed to increase the probability of successful treatment by increasing the probability that the partner receives treatment, reducing the time to treatment of the partner and reducing the risk of reinfection in the index case and partner. Scenarios 1 and 2 (see Table 14) involve screening for chlamydia only with the addition of standard patient referral (scenario 1) or APT (scenario 2). In scenarios 3 (routine testing of partners for chlamydia and gonorrhoea) and 4 (APT for chlamydia and gonorrhoea), all individuals who receive screening are tested for chlamydia and gonorrhoea and, if positive, treated. If the test result yields at least one positive test, the most recent partners are also treated for chlamydia and gonorrhoea.

Figures 20a–d show the effects of the scenarios in the models with and without interactions and in the population overall and those in the highest risk class. With a screening interval of 5 years (screening rate 0.2 per year), scenario 1 (screening plus standard patient referral) can reduce the prevalence of chlamydia in the general population by about one-third (see Figure 20a). This is consistent with the expected effect of chlamydia screening (see Modelling the transmission dynamics of C. trachomatis and Individual- and population-level effects of partner notification for C. trachomatis). Scenario 2 (screening plus APT) results in slightly lower chlamydia prevalence than scenario 1. Interestingly, in the model where the infections co-interact, we observe a small reduction in gonorrhoea prevalence, although treatment (with or without testing) is performed only for chlamydia (see Figure 20c and d). Hence, interactions between chlamydia and gonorrhoea could potentially help to indirectly reduce the prevalence of gonorrhoea in a population that is tested and treated for chlamydia.

FIGURE 20.

Effects of screening and partner notification for chlamydia (a and b) and gonorrhoea (c and d), assuming that the average screening interval for all individuals is 5 years. The prevalence rates of four different scenarios are contrasted with those in absence of any intervention. Strategy 1, routine testing for chlamydia; strategy 2, APT for chlamydia; strategy 3, routine testing for chlamydia and gonorrhoea; strategy 4, APT for chlamydia and gonorrhoea. Overall (a and c): general population of 16- to 29-year-olds. High risk (b and d): individuals with 10 new heterosexual partners per year (1.5% of the general population).

Routine testing for chlamydia and gonorrhoea (scenarios 3 and 4) further reduces the prevalence of gonorrhoea (see Figure 20c and d). Cotesting for gonorrhoea has a negligible impact on reducing the prevalence of chlamydia in the model where we assume interactions between the two infections (see Figure 20a and b), due to the low prevalence of gonorrhoea. APT (scenarios 2 and 4) always results in lower prevalence rates of chlamydia and gonorrhoea compared with standard patient referral (scenarios 1 and 3), particularly for chlamydia but also for gonorrhoea in the case where the two infections interact. This exemplifies the improved clinical effectiveness of APT over standard patient referral for bacterial STIs. Figure 21 shows the results of the same scenarios, assuming that individuals receive screening at intervals of 2 years. The pattern of results is the same but effects are stronger.

FIGURE 21.

Effects of screening and partner notification for chlamydia (a and b) and gonorrhoea (c and d), assuming that the average screening interval for all individuals is 2 years. The prevalence rates of four different scenarios are contrasted with those in absence of any intervention. Strategy 1, routine testing for chlamydia; strategy 2, APT for chlamydia; strategy 3, routine testing for chlamydia and gonorrhoea; strategy 4, APT for chlamydia and gonorrhoea. Overall (a and c): general population of 16- to 29-year-olds. High risk (b and d): individuals with 10 new heterosexual partners per year (1.5% of the general population).

It is surprising that the prevalence of gonorrhoea cannot be lowered through widespread screening at intervals of 2 years (screening rate 0.5 per year). This goes counter to the usual observation that for STIs with very low endemic prevalence rates, only minor interventions can result in extinction. In our model, however, about one imported case per month is co-infected with chlamydia and gonorrhoea (in a total population of 10,000 individuals), which keeps gonorrhoea infections at an endemic level.

Gonorrhoea prevalence can be subjected to strong fluctuations with occasional outbreaks in a population. Partner notification strategies that provide treatment without STI testing might facilitate outbreaks. We calculated the yearly prevalence of gonorrhoea in the total population, and defined an outbreak when the prevalence exceeded 0.5%, which was the baseline prevalence in the model population (Figure 22). The frequency of outbreaks increased very slightly, compared with no intervention, only in the model that assumed no interaction between infections. In the model that incorporated an interaction, both APT and standard partner notification strategies resulted in a reduced frequency of gonorrhoea outbreaks. The greatest reduction in gonorrhoea outbreaks was found in the model in which the two infections interact and in which individuals are tested for chlamydia and gonorrhoea using APT (scenario 4).

FIGURE 22.

Outbreak frequency of gonorrhoea. (a) 5-year screening interval; (b) 2-year screening interval. Gonorrhoea prevalence is measured once a year and an outbreak is defined when the observed annual prevalence in the general population exceeds 0.5%. The prevalence rates of four different scenarios are contrasted to those in absence of any intervention. Strategy 1, routine testing for chlamydia; strategy 2, APT for chlamydia; strategy 3, routine testing for chlamydia and gonorrhoea; strategy 4, APT for chlamydia and gonorrhoea. Overall: general population of 16- to 29-year-olds. High risk: individuals with 10 new heterosexual partners per year (1.5% of the general population).

Reinfection of index cases

The transmission dynamic model assumes that sexual partnerships are instantaneous events. Therefore, we cannot directly investigate the outcome of reinfection of treated index cases by their untreated partners. Instead, we use a simplified model that investigates rates of reinfection with chlamydia and gonorrhoea only. We generated 1000 parameter sets by randomly drawing from the following distributions: the probability that index cases are in an ongoing partnership is uniformly distributed between 0 and 1; the probability that partners are infected with chlamydia is uniformly distributed between 60% and 70%;7 the probability that partners are infected with gonorrhoea is uniformly distributed between 70% and 80%;6 the frequency of sex acts is uniformly distributed between once a day and once a week; the per-sex act transmission probability is uniformly distributed between 6% and 16.7%;48 the per-sex act transmission probability for gonorrhoea is assumed to be twice that of chlamydia; and the duration of sexual partnerships is uniformly distributed between 1 week and 6 months. 48

We calculated the reinfection rates for chlamydia and gonorrhoea for different delays of partner referral (Figure 23). Reducing the delay between index case and partner treatment from 14 days111 to 1 or 3 days20 substantially reduces the rates of reinfection. For chlamydia, the median reinfection rates from the 1000 parameter combinations are 7.6%, 2.3% and 0.8% for decreasing delays of partner referral. For gonorrhoea, the corresponding rates are 14.6%, 5.4% and 2.1%.

FIGURE 23.

Reinfection rates of treated index cases by their untreated partners. (a) Chlamydia; and (b) gonorrhoea. The figures show box plots of 1000 different parameter combinations.

Discussion

There is some evidence from epidemiological and animal studies suggesting an interaction between chlamydia and gonorrhoea infections. 6,107,108 Observed co-infection rates cannot be obtained in mathematical models that assume that gonorrhoea and chlamydia are transmitted independently. The results from a model show that a potential interaction between the two organisms can influence the effect of testing and treating index cases and their partners for chlamydia and gonorrhoea. Overall, we found that cotesting and cotreatment for gonorrhoea can lower its prevalence in the general population as well as in individuals with high sexual activity. However, the level of reduction is not as pronounced as for chlamydia, which is partly explained by the ongoing import of new infected cases. In terms of partner notification, APT reduced infection prevalence more than standard patient referral. In addition, shortening the delay to partner treatment reduced rates of reinfection of treated index cases from their untreated partners.

To our knowledge, this study presents the first mathematical model of chlamydia and gonorrhoea infection that describes the observed co-infection rates. While biological co-interactions between chlamydia and gonorrhoea are a plausible mechanism, there might be other explanations that could explain the high rates of co-infection. For example, the specific group of individuals who are co-infected with chlamydia and gonorrhoea could be a subpopulation that has both frequent sex partner change rates and low usage of condoms. Chlamydia infections might be able to reach a very high prevalence in this kind of population. Also, highly assortative mixing by certain behavioural or demographic characteristics could increase rates of co-infection. These and other mechanisms should be further investigated with mathematical models to better understand the observed co-infection rates of chlamydia and gonorrhoea but also of other STIs such as syphilis and HIV.

We kept our modelling approach deliberately simple so that we could study general principles of the effect of interactions between chlamydia and gonorrhoea on the impact of interventions. First, we considered a general heterosexual population of 16- to 29-year-olds. Gonorrhoea infections and co-infection rates are higher in men who have sex with men than in the general heterosexual population, but should be studied in a separate model. Second, we ignored age structure in the level of sexual activity, but we believe that the four risk groups describe the heterogeneity in sexual activity and the transmission of chlamydia and gonorrhoea reasonably well. The model could be developed further by adding different ethnic or social groups among whom chlamydia and gonorrhoea rates vary. 43 Third, we ignored differences in disease-specific parameters among women and men; biological interactions between the two organisms could differ between the two sexes. These additional complexities will be important to consider in future modelling studies of the nature of chlamydia and gonorrhoea co-infection.

The findings from this study suggest that, if there is a strong biological interaction between chlamydia and gonorrhoea, treating chlamydia infections should help to reduce the overall prevalence of gonorrhoea. Although an association between trends of chlamydia prevalence and gonorrhoea prevalence has not been observed empirically, it might be possible to investigate whether or not gonorrhoea incidence rates have declined in populations that were regularly tested for chlamydia. The model results suggest that APT should be favoured over standard patient referral because it results in lower infection prevalence and can substantially reduce reinfection of index cases from their untreated partners. However, future work is required to obtain better estimates of the probabilities of successful treatment of index cases and their partners, given a specific partner notification technology. Furthermore, the risks of not testing for HIV in partners treated via APT should be explored in modelling studies. Finally, we found that a strategy of dual testing and treating in the general heterosexual population had a modest impact on reducing the prevalence of gonorrhoea. The model did not investigate the numbers of false positive test results, however, which might counteract the benefits, particularly in low prevalence settings.

Introduction

This section outlines the challenges of measuring the cost-effectiveness of partner notification using the outcome of cost per QALY. In a systematic review of economic evaluations of interventions to prevent chlamydia infections,112 we identified three studies published up to 2004 that evaluated partner notification. One presented the outcome as major outcomes averted113 and two reported short-term outcomes. 58,114 None reported cost per QALY.

Objective

To describe alternative approaches that have been used to assess and value the quality of life associated with the reproductive sequelae of chlamydia infection in women.

Methods

We conducted a systematic review of published literature. We followed a two-stage process for screening the results of electronic database searches, as in the earlier review. 112

Results

The electronic database search identified 3534 published studies. From these, 1178 studies were duplicates. Titles and abstracts were reviewed for the remaining 2356 studies. Hand searching of bibliographies from published literature identified one additional study. Seventeen studies were considered potentially relevant. Of these, eight studies were excluded because they did not discuss the instruments and techniques available for eliciting the value of health states for avoiding chlamydia. No studies were excluded on the grounds of language. Figure 24 shows the flow of papers retrieved and included.

FIGURE 24.

Flow chart of included studies.

Discussion

This systematic review found only one study that provides QALY weights for reproductive tract complications of chlamydia infection. No studies measuring HRQL or health state valuations among women in the UK or in the rest of Europe were found.

The main strength of this study is the comprehensive literature search. We do not think that our search strategy missed major studies. Nevertheless, the only study that reported QALY weights was published in a book chapter and was identified from reference lists of other studies and not from an electronic database search. Weaknesses of the study relate to difficulties in drawing conclusions from the small number of studies retrieved.

To our knowledge, this is the first systematic review to give a critical appraisal of methods used to elicit HRQL information for the economic evaluation of interventions for chlamydia prevention. The studies included in this review demonstrate many of the methodological debates about health state valuation. 125 We reviewed studies that collected data from patients, healthy adults and experts. Torrance et al. 127 suggest that judgements made by experts may be a starting point and are quick to collect. Experts might, however, have different views from those of patients. A major weakness of the IoM study was the lack of detail describing the elicitation process and the size, make-up and selection process of the expert panel, and so the potential for bias in the results cannot be assessed. Current NICE guidelines128 state that HRQL should be elicited directly from patients with the condition and the utilities should be based on public preferences using the European Quality of Life-5 Dimensions.

Partner notification works by reducing the duration of a STI, which is already a temporary health state and might be asymptomatic. Valuing temporary health states is challenging, particularly when most HRQL instruments have been designed to measure health states resulting from chronic or terminal diseases such as arthritis and cancer. This review did not identify any studies explicitly valuing the temporary health state of chlamydia infection in the absence of complications, although the two-stage approach used by Smith et al. 118 appears to have implicitly addressed this. Although STIs do cause physical and emotional damage, complications typically occur in the future and can also be temporary. Costs and benefits in the future are valued less than those occurring in the present, and so preventative interventions for curable STIs are less likely to appear cost-effective than those for ongoing chronic conditions.

Chronic pelvic pain might be easier to assess using currently available instruments. The valuations that we reviewed suggested a substantial detrimental impact on quality of life. 116 Scenarios describing chronic pelvic pain suggest, however, that it is usually very severe and long lasting. Evidence from longitudinal studies117 and from TTO studies118 suggests, however, that both the duration and the severity might have been overestimated. An additional challenge for estimating QALYs is that the fraction of chronic pelvic pain attributable to chlamydia and other STIs is unknown.

This review has implications for future research. HRQL data for chlamydia and its complications should be collected in the UK using appropriate instruments and study populations. Better QALY estimates for outcomes associated with chlamydia are also needed. The methods used to derive the IoM estimates are not transparent and do not meet current NICE guidelines. Based on the results of this systematic review, it was not possible to draw a firm conclusion about the most appropriate techniques for measuring HRQL and for valuing outcomes associated with chlamydia. We did not conduct a formal cost-effectiveness evaluation of partner notification because of the unresolved methodological challenges and the absence of appropriate evidence.

Introduction

Partner notification for chlamydia in community health service settings should take place as often as in genitourinary medicine clinics in England because of the large number of cases diagnosed through the NCSP. The standards for partner notification outcomes in the NCSP are the same as in UK genitourinary medicine clinics (see Clinical effectiveness of partner notification technologies: systematic review): 0.6 partners per index confirmed treated (0.4 in inner cities). 45 The number of partners treated per index, however, has fallen from a high of > 0.55 in 2006–7 to just under 0.4 in 2010–11. There is no standard protocol for providing partner notification services in community settings such as general practice and community Contraceptive And Sexual Health (CASH) services (also known as family planning clinics). Different areas operate with very different models of partner notification, ranging from minimal involvement of health-care professionals to active follow-up of index cases and their partners in order to confirm that partners have been treated. Chlamydia Screening Offices (CSOs), which are part of NCSP infrastructure in some areas, can provide more intensive support.

We have shown that chlamydia positivity rates of sexual partners of index cases with chlamydia are 5 to 10 times higher than in the general population (see Individual- and population-level effects of partner notification for C. trachomatis). Partner notification is therefore efficient for case finding. The UK National Audit Office published estimates of the cost of a chlamydia-screening test in 2009, using NCSP data from PCTs; these were updated in the chlamydia cost-guidance initiative. The National Audit Office reported an average cost per screen of £45 in 2009, although a wide range of costs was reported in different PCTs. 129 There were no published data on the cost of partner notification.

The coverage of chlamydia screening in all settings in 2011–12 was > 30% in women in all PCTs but < 20% in men. 130 Improving overall screening coverage will therefore depend on increasing the percentage of young men being screened. Given the renewed focus of the NCSP on effective management of positive individuals and their partners, monitored by the diagnostic rate indicator,130 information about the relative costs and cost-effectiveness of expanding screening coverage and partner notification services in the NCSP is needed.

Objectives

To improve estimates of the relative costs of opportunistic chlamydia screening and partner notification in the context of the English NCSP.

Methods

We used a simple spreadsheet model to investigate the costs of screening and partner notification for chlamydia. The spreadsheet was developed to help local areas explore their own situation within the context of the national picture and help them plan service provision for chlamydia screening and partner notification activities, based on NCSP data for 2008–9. 31 The model is freely available (www.bmj.com/content/342/bmj.c7250?tab=related#datasupp). The spreadsheet model 2008–9 estimates cost per individual tested; cost per positive diagnosis; total cost of screening; number screened; number infected; and sex ratio of those tested and treated.

For this project, we updated the spreadsheet using NCSP data for 2010–11. We then examined a range of scenarios, based on parameter estimates derived from other studies in this project (see Partner notification outcomes for chlamydia in UK genitourinary medicine clinics; Estimating the likely public health impact of partner notification for a clinical service: an evidence-based algorithm; and Modelling the transmission dynamics of C. trachomatis). For each scenario we compared the cost-effectiveness of two interventions using the cost per positive chlamydia case diagnosed: increased partner notification efficacy and increased coverage of primary chlamydia screening in men.

Scenario analysis

Table 19 shows additional scenarios, using a range of estimates for partner notification efficacy (number of partners treated per index; values used 0.4, 0.6, 0.9, 1.2); percentage of partners infected (0.5, 0.65, 0.7); and screening coverage [proportion of target population screened in year (men 8%, women 24%; both 20%, both 30%)].

Any combination of parameters can be fed into the model: a small selection of parameter combinations are presented here as examples. In scenarios 1–7 the underlying screening coverage remains constant, and as the partner notification efficacy or partner positivity increase, the total number of infections treated increases. Hence, the total diagnosis rate increases and the cost per positive decreases. Increasing either partner notification efficacy or partner positivity reduces the cost per positive case identified. As partner positivity has been shown in modelling studies to be insensitive to the underlying prevalence, efforts to improve partner notification efficacy are likely to result in greater cost-effectiveness (i.e. reduced cost per positive case identified regardless of other infection dynamics).

In general this model suggests that it is more cost-efficient to increase partner notification for chlamydia positive patients rather than screening greater numbers of low-risk individuals. For example, increasing the mean partner notification efficacy from 0.4 to 0.9 (scenarios 5 and 3) would be expected to increase the overall diagnosis rate from 1575 to 1982: a 26% increase.

Discussion

The spreadsheet model suggests that increasing the efficacy of partner notification costs less per new positive case diagnosed than increasing the coverage of chlamydia screening in men or in both sexes. The conclusions are unchanged when using updated data from the NCSP for 2010–11 and applying a range of parameter estimates for different components of the model.

A limitation of this simple static model is that it is not possible to estimate long-term cost-effectiveness within this model framework; that would require a transmission dynamic model. We took a conservative approach to determining the relative cost-effectiveness of partner notification by using the highest estimated cost of providing partner notification services. We also intentionally excluded the effects of partner notification on preventing reinfection in treated index cases. Including lower costs of partner notification and the costs averted by preventing reinfection would make partner notification more cost-effective. We took the cost perspective of the health service and restricted ourselves to consideration of the costs of screening and partner notification, and excluded patient costs, the costs of reinfection and the cost of complications arising from the initial infection.

Of note, as long as the positivity of those screened opportunistically remains proportionately much lower than that of partners of infected persons (e.g. ∼5% screen positivity vs. ∼30–60% partner positivity), investing in effective partner notification is a good use of resources. A modelling study has shown that the proportion of partners of an index case who are infected with chlamydia is consistently about 30%. This proportion is insensitive to the underlying population or network prevalence. In contrast, the positivity in those screened is sensitive to the underlying prevalence in the population, changing in line with changes to underlying prevalence.

As the number of index cases is small compared with the number screened, the cost of partner notification is a fraction of the total cost of screening. This remains true even if we make pessimistic assumptions about the cost of partner notification. In fact, partner notification can probably be done effectively at very low cost if most notifications are managed through patient referral (with appropriate motivation and support) plus effective use of web and text-messaging technologies to provide results and information to patients and their partners as well as efficient monitoring and follow-up of outcomes.

There are, however, implications and challenges for services implementing the NCSP. There can be confusion about the definition of partner notification outcome measures and denominator populations. Better information and support in providing these data would improve the quality of monitoring outcomes of partner notification in the NCSP.

In Chapter 3 (see Individual- and population-level effects of partner notification for C. trachomatis), we showed that the most recent partner is the most likely to be infected and the probability declines with time since end of partnership. In addition, we have shown that infected casual or irregular partners are more likely to transmit the infection to other people (see Chapter 2, Estimating the likely public health impact of partner notification for a clinical service: an evidence-based algorithm). 9 Therefore, even if benefit to the index case of treating past partners is limited, the wider population gains more benefit. In this model, we do not distinguish between types of partnership, but it seems likely that increasing the effectiveness of partner notification will result in a reduction in partner positivity as more non-regular partners are included.

Improving the diagnosis rate will be a key indicator of programme performance and monitoring. We demonstrate that significant improvements can be made to this outcome measure through improved case management and partner management.

Clinical effectiveness of partner notification technologies: systematic review

Adel Ferreira, Taryn Young, Catherine Mathews and Moleen Zunza are co-authors of the Cochrane Collaboration systematic review of strategies for partner notification. 44

Partner notification outcomes for chlamydia in UK genitourinary medicine clinics

Hugo McClean and Chris A Carne are co-authors of the analysis of clinical audit data about partner notification outcomes for chlamydia. 45

Estimating the likely public health impact of partner notification for a clinical service: an evidence-based algorithm

Catherine R H Aicken, M Gary Brook and Claudia S Estcourt are co-authors on the evidence-based algorithm for estimating the public health impact of partner notification. 9

Modelling the transmission dynamics of Chlamydia trachomatis

Janneke C M Heijne and Adrian Roellin are co-authors on the articles describing the estimate of C. trachomatis infectious duration46 and investigating the individual and population-level effects of partner notification. 49

Individual- and population-level effects of partner notification for Chlamydia trachomatis

Janneke C M Heijne is a co-author on the article estimating the transmissibility of C. trachomatis. 47

Outcomes for the economic evaluation of partner notification technologies: systematic review

C Pswarayi contributed to the systematic review of quality-of-life studies. 136

Costs and cost-effectiveness of different strategies for chlamydia screening and partner notification: an economic and mathematical modelling study

Elisabeth Adams, Arabella Grant, John Macleod, Jan Clarke and Paddy Horner are co-authors to the study of the costs and cost-effectiveness of chlamydia screening and partner notification strategies. 31

Editorial assistance

Kali Tal edited several of the chapters for language, style and consistency.

Literature reviews and other secondary data

Mathematical modelling

We will use three existing UK-developed models of chlamydia and/or gonorrhoea transmission. Our modelling approach takes into account uncertainty in a) parameter estimates and b) model structure in existing and is informed by methods described by Brisson and Edmunds9 and Koopman. 10 We are currently pioneering this technique in our ongoing work to improve the value of modelling studies for public health decision-making about chlamydia screening. 11 Given the variability in outputs (Figure 1), the use of multiple models will improve the robustness of the inferences that can be drawn by allowing differences between models to be explained.

Model comparison: We will begin by comparing in detail the structures, baseline assumptions and predictions of the existing models using features common to all models, for example, predicted chlamydia prevalence in heterosexuals by age and sex following epidemiological treatment and testing for 100%, 50% and 20% of current partners. This will allow a ‘diagnosis’ of parameter and model differences that are likely to result in differences in outputs. Each model offers different advantages and disadvantages, which are relevant to the project’s objectives. For example, the ClaSS project model models the incidence of infertility dynamically over time, which is important for estimating the long term impact of partner testing and treatment technologies. The first round of model comparison found that none of the available individual based models was suitable, and that improved estimates were needed for some parameters. Models specific to each study question were then developed. The Imperial College and HPA models have already been developed to consider both chlamydia and gonorrhoea, expediting investigation of the effects of co-infection with two key infections. Each team will then program their model to give a population that includes the possibility of studying multiple infections, hetero and homosexual partnerships, ethnic heterogeneity, genitourinary medicine and primary care treatment settings, partnership network structure and complications. The Imperial College model could not be developed further, owing to changes in staff.

Clinical outcomes: The main modelling work will focus on determining the most plausible internally and externally consistent estimates of the outcomes important of public health decision making: total and age and sex specific transmissions and complications prevented and, for the urban mixed population, outcomes in both black and white ethnic groups. For each selected infection and partner testing scenario, we will follow a similar process to determine the impact on both the index infection and co-infections with respect to the effects of, a) seeking care for one infection and having others diagnosed, b) co-treatment, and c) lost opportunities to diagnose and treat other infections. For example, for chlamydia we would examine outcomes for, a) base case (epidemiological treatment with testing for index and co-infections), b) the index patient’s current partner treated for chlamydia on the same day but not tested for co-infections, c) current partner being tested for chlamydia but not treated until results are known. We will conduct an iterative process of model comparison, diagnosis, re-parameterisation and programming. At each stage differences between model outputs will be discussed, and reasons explored. We will test alternative assumptions about parameter values, for example, the uptake of different strategies, in sensitivity analyses. We will explore differences due to model structure by modifying one model to reproduce the structure of the one (or ones) from which it differs. If final model outputs converge to produce similar predictions, the final output will be the average from all three models. If final outputs remain different, with no available empirical data to help determine the most plausible outputs, we will present a low, middle and upper estimate. The numbers of infections and complications averted will be used in the economic evaluation. Models specific to each study question were developed. Complications were not modelled, owing to lack of time.

Economic evaluation and health state valuation

We will use published sources to estimate the costs of antibiotic treatment, face-to-face consultation, telephone consultation, hospital and/or outpatient treatment for tubal infertility, ectopic pregnancy, pelvic inflammatory disease, HIV infection at 2009 Sterling prices. 1^;12^;13

Date: 22 March 2011, 29 January 2012 and 31 August 2012