REACH: a mixed-methods study to investigate the measurement, prediction and improvement of retention and engagement in outpatient HIV care

Outpatient attendance patterns among people living with HIV

The UK CHIC collates routine data relating to the clinical care and treatment of PLWH aged ≥ 16 years across many of the UK’s largest HIV services (the data set utilised for the present analyses included data from 15 HIV clinics) since 1 January 1996. The UK CHIC records incorporate additional mortality data provided by Public Health England.

We explored the use of group-based trajectory modelling for identifying clusters of individuals following similar, distinctive progressions of attendance over age or time. 39–41 These methods were applied to the UK CHIC data set in order to distinguish distinct broad groupings of individuals according to their pattern of engagement with the aim of identifying a unique, statistical snapshot of the key characteristics of this complex population. We collaborated with Dr Tracy Glass at the Basel Institute for Clinical Epidemiology and Biostatistics, who has utilised a similar technique to explore and describe patient adherence to ART in the Swiss HIV Cohort Study. 42 Once the trajectories are identified, logistic regression analysis can then be used to predict the probability of an individual being in a particular group according to a particular set of risk factors.

Group-based trajectory analysis is usually applied to a fixed period of time during which individuals are exposed to a condition or data are collected at fixed points in time and a measure of their response can be plotted over time. We therefore developed an EIC algorithm in order to define whether patients were in care or out of care for each month of follow-up.

Development of engagement in care algorithm

As we know that frequency of HIV clinic attendance is related to changes in treatment and health status, our aim was to incorporate clinical factors into our new measure of retention in HIV care.

In the absence of complete data on clinic attendances in the UK CHIC data set, we used CD4 counts, viral loads, haemoglobin measures and ART treatment start or switch dates as markers of clinic attendance. Patients often return for repeat laboratory tests over a short period of time to confirm unexpected findings, resulting in clusters of attendances around a single ‘index’ date. As we did not want to consider each attendance within a cluster as an independent visit and wanted to be able to plot whether patients were in care or out of care for each month of follow-up, we grouped attendances into ‘care episodes’, defined as months (period of 30.4 days since entry into the study) where at least one visit occurred. For each care episode, we established the lowest CD4 count measured in that month (and the change from the previous value), the highest HIV viral load (and the status of this measurement relative to other consecutive values) and the patient’s treatment status.

In order to decide which key clinical factors should be incorporated the EIC algorithm, semistructured, face-to-face interviews were conducted with eight HIV physicians (from August 2013 to February 2014) who had a range of clinical experience and were selected from five of the HIV outpatient clinics where we planned to conduct our primary data collection. We asked physicians to tell us when they planned to see each of their last 10 patients again (number of weeks/months) and why.

A total of 73 patients were discussed in these interviews. One patient was excluded from the analysis because the time of their next appointment was dependent on awaited test results. The time of the next scheduled appointment was missing for another patient and not available for a further five patients who had not attended their last scheduled appointment at the time of the physician interviews. We conducted a content analysis of these qualitative data. For each patient, we noted the time to the next scheduled appointment and the key reason given by their physician. We then identified factors under which to code the key reasons.

The time of the next scheduled appointment in the 66 patients included in the analysis ranged from 1 week to 6 months, with a median of 3 months. Five factors were identified from the content analysis of the interview data as instrumental to the timing of the next scheduled appointment. The first factor can be summarised as ‘routine’ where patients were stable and required routine follow-up. These appointments were mostly arranged for 4–6 months after the previous visit. A routine follow-up appointment was arranged for 3 months’ time for one pregnant woman. Physicians talked about how they extended routine visits to every 6 months when patients were well and stable, both on treatment and in their psychosocial circumstances. The second factor is summarised as ‘virological’ where the next appointment was based on change in viral load (uncontrolled or virological breakthrough). Changes in viral load brought the next scheduled appointment forward to 1–2 months after the last. The third factor, ‘treatment’, described where the next appointment was related to starting ART or changing an existing ART regimen. Patients were given a next appointment date between 2 and 12 weeks later, depending on the treatment start date and virological response or when treatment was planned to start. The fourth and fifth factors were ‘psychosocial’ where mental health or psychosocial issues were identified as instrumental and ‘physical comorbidities’ where a range of physical comorbidities were given as the key reason for the timing of the next appointment. Follow-up appointments of between 1 week and 4 months were arranged depending on psychosocial issues (from specific concerns about mental health to more general needs for social support) and comorbidities: both of which required earlier follow-up even when patients were otherwise stable on treatment.

The data from the physician interviews were used as the basis for developing the EIC algorithm. Although psychosocial well-being and comorbidities were key factors in determining the expected time between patients’ visits, data on these variables are not captured in the UK CHIC data set. We therefore used the clinical data that were available (HIV diagnosis, AIDS diagnosis, treatment start dates, CD4 count and viral load) to determine the patient’s treatment and health status. These data were then used to estimate the expected time to the next scheduled care episode, in accordance with the data collected in the physician interviews. Table 1 shows when the next care episode was expected, according to the specified conditions. Using the EIC algorithm, we cannot consider follow-up after the patient’s last reported care date and it therefore focuses on intermittent periods of disengagement rather than LTFU. It should furthermore be noted that treatment guidelines introduced in the UK in October 2015 (after our physician interviews had been conducted) recommend starting ART irrespective of CD4 count. 43 This should be incorporated into the EIC algorithm when applied to EIC after October 2015.

TABLE 1 - Conditions associated with the expected time of next scheduled care episode

If more than one condition applies at time of care episode, next care episode is expected within least number of months associated with those conditions.

Blips are defined as having a viral load of between 50 and 200 copies/ml following a previous viral load of < 50 copies/ml.

Adapted from Howarth et al. 44 This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Conditions at time of initial care episodea	Next care episode expected within
Within 1 month of HIV diagnosis	2 months
AIDS diagnosis	2 months
Started ART	2 months
Started new combination ART regimen	2 months
Not on ART
CD4 ≤ 350 cells/mm3, any drop in CD4	2 months
CD4 ≤ 350 cells/mm3, no drop in CD4	4 months
CD4 = 351–499 cells/mm3	4 months
CD4 ≥ 500 cells/mm3, CD4 drop ≥ 100 cells/mm3	4 months
CD4 ≥ 500 cells/mm3, CD4 drop < 100 cells/mm3, viral load ≥ 100,000 copies/ml	4 months
CD4 ≥ 500 cells/mm3, CD4 drop < 100 cells/mm3, viral load < 100,000 copies/ml	6 months
Already started ART
Viral load > 200 copies/ml	2 months
Viral load = 51–200 copies/ml, does not appear to be a blipb	2 months
Viral load = 51–200 copies/ml, appears to be a blip	4 months
Viral load ≤ 50 copies/ml, CD4 ≤ 200 cells/mm3	4 months
Viral load ≤ 50 copies/ml, CD4 > 200 cells/mm3	6 months

The EIC algorithm gives the shortest expected gap between care episodes at 2 months. This is to allow for the fact that clinic visits might occur at any point during the month or care episode into which they are grouped. If the patient was within 1 month of diagnosis, had an AIDS diagnosis, started ART or changed ART at the initial care episode, the next care episode was expected within 2 months. If the patient was not on ART at the initial care episode, the next care episode was expected within 2–6 months, depending mainly on CD4 count. If the patient had started ART, it was expected within 2–6 months, depending on viral load. We used 6 months as the maximum time between visits, as described in the physician interviews. If more than one condition applied at the time of the initial care episode, the next care episode was expected within the least number of months associated with those conditions.

Figure 1 shows an example of how the EIC algorithm is applied to an individual case. The date of the next observed care episode determines whether or not the patient has attended before or after the expected date, and each patient-month is then classified as being in care (where it occurs on or before the time of the next expected care episode) or out of care (where it occurs after the time of the next expected care episode) accordingly. This example begins in July 2005 when the patient was not pregnant, not recently diagnosed, did not have an AIDS diagnosis and was not on ART. Because she was not on ART, we are interested in her CD4 count – which was < 350 cells/mm³ and which had dropped (although that is not shown here), so we expect to see her again within 2 months. However, she did not reattend until 6 months later. Thus, July and August are defined as being in care (blue shading), but September to December are out of care (green shading). At the visit in January 2006, her CD4 count had continued to drop, so we expect to see her again within 2 months. She reattended in March 2006 so January and February are defined as being in care. In March 2006 she started ART, so we expect to see her within 2 months and she comes back within a month, remaining in care.

FIGURE 1.

Measurement of engagement in HIV care applied to an individual care.

Association with clinical outcomes

In view of the limitations of using the findings from our group-based trajectory analysis in further analysis to examine the associated clinical outcomes, we decided to use the proportion of time in care as a more straightforward and fitting measure of engagement in HIV care for further analysis. The individual case shown in Figure 1, for example, was out of care for 4 of her 10 months’ follow-up and in care for 6 of her 10 months’ follow-up = 60% of months. The EIC measure, furthermore, provides an alternative to the idea that disengagement from care requires an individual to be LTFU for a specified period of time which does not lend itself to quantifying the typical length of disengagement or proportion of patients that re-engage with care, as described in our protocol.

In the following analysis, all patients who attended a participating UK CHIC clinic on two or more occasions between 1 January 2000 and 31 December 2012 were included. The proportion of months in which patients were engaged in HIV care was calculated overall and for patient subgroups defined by gender, age group (< 25, 25–45 and > 45 years), ethnic group (white, black African, other and unknown), mode of HIV acquisition (sex between men, sex between men and women, injection drug use and other/unknown), currently on ART (yes, no), nadir and current CD4 count (both classified as < 200, 200–349 and ≥ 350 cells/mm³), participating clinic, calendar year (2000–3, 2004–7, 2008–12) and time since entry in the study (< 1, 1–5, 5–10 and > 10 years). Each patient-month was then treated as a separate entry in a multivariable logistic regression model with the aim of identifying demographic and clinical factors associated with that month being in care. The analyses were performed using PROC GENMOD in SAS version 9.4 (SAS Institute Inc., Cary, NC, USA) with generalised estimating equations used to take account of the repeated entries within each individual patient.

Cox models were used to assess the association between mortality and (i) the cumulative proportion of months a person had been in care; and (ii) the cumulative proportion of months a person was in care prior to ART. For analysis (i), patient follow-up started on the date of entry to the UK CHIC study and ended at the date of death or 6 months after the patient’s last clinic attendance, whichever occurred first. Each person’s total period of follow-up was split into consecutive monthly intervals (as described above) and the cumulative proportion of previous months a patient had been in care at the start of the month was calculated for each month of follow-up. In this way, we were able to include the EIC covariate as a continuous time-updated covariate with each person’s EIC evolving over time in the model. One of the main limitations of analyses that investigate associations between EIC and mortality, particularly where a time-updated covariate is used, is the potential for reverse causality whereby those who are sickest may attend for care more frequently in the months leading up to death. This may make it appear that higher levels of EIC are associated with an increased risk of mortality. In order to reduce the potential for this to bias our analyses, all measures of EIC were lagged by 12 months to separate the assessment of EIC and the mortality outcome by a period of 1 year. At any given time over follow-up (i.e. at the start of each patient-month that is included in the model), therefore, the lagged value of EIC that was entered into the model was the value that was available 12 months earlier. Thus, our estimate of the relative hazard (RH) associated with EIC will provide a description of the ability of our EIC measure to predict mortality events that occur at least 1 year into the future. Note that, by definition, this approach will necessarily restrict analyses to those who had attended clinic for > 1 year, as individuals who die within the first 12 months of follow-up will not contribute any EIC values to the model and will therefore be excluded. In our primary analyses, we adjusted for the demographic factors of age, year, gender, mode of acquisition and ethnic group (all fixed covariates). This was followed by additional adjustment for receipt of ART as a binary time-updated covariate. We next adjusted for the latest CD4 count, as a continuous time-updated covariate and lagged by 12 months, to investigate whether or not any association seen was explained by the fact that those with the lowest EIC values already had lower CD4 counts at the time of measurement of EIC. Finally, we adjusted for the unlagged values of CD4 – this analysis explored whether or not any residual association between EIC and mortality was mediated by lower CD4 counts over the following 12 months.

For analysis (ii), we calculated the EIC prior to ART in the subset of patients who initiated ART and who had been under follow-up at the clinic for at least 1 year prior to ART start (to ensure that the estimate of EIC was based on sufficient patient-months to provide a robust estimate). To explore the potential for confounding, patients were first stratified into six groups based on their pre-ART EIC. The groups (< 50%, 50–69.9%, 70–79.9%, 80–89.9%, 90–99.9% and 100%) were chosen primarily for ease of clinical interpretation and to ensure that each group was of sufficient size to permit robust analyses. Associations between the pre-ART EIC value and the various demographic and clinical factors at ART start were then identified. Next, Cox models considered the association between EIC (as a fixed baseline covariate) and mortality after ART initiation; follow-up started at ART initiation and ended at 6 months after the patient’s last visit or death – whichever of the two occurred first. We first adjusted for age, year, gender, mode of acquisition and ethnic group, then type of ART received [protease inhibitor (PI)-based regimen, non-nucleoside reverse-transcriptase inhibitor (NNRTI)-based regimen, other regimen (including those on both a PI and NNRTI)], then the CD4 count and viral load at ART start (fixed covariates) and then, finally, the latest CD4 count and viral load (as time-updated covariates) measured after ART start. As before, these last analyses investigate whether or not any associations between pre-ART EIC and post-ART mortality can be explained by poorer CD4/viral load responses on ART.

Association with laboratory tests and costs

For our health economic analysis, we planned to measure the impact of disengagement on HIV management costs. As noted in our study protocol, such an analysis was likely to be difficult as it would be limited by the lack of NHS resource utilisation data captured in UK CHIC. Given the limited data for some but not all clinics on outpatient clinic attendances and inpatient stays we therefore limited our study to a descriptive analysis to selected laboratory test costs, which are reliably collected in UK CHIC across participating clinics.

Using the ART data set from analysis (ii), we calculated the total number of laboratory tests performed within each pre-ART EIC group within each 3-month period of follow-up after ART start until the end of the fourth year after starting ART. We included the following tests:

• CD4

• HIV viral load

• liver function tests

• cholesterol

• high-density lipoprotein

• low-density lipoprotein

• triglycerides

• full blood count

• urea

• creatinine

• glucose

• bone health.

Liver function tests included one or more of the following blood tests: alkaline phosphatase, alanine transaminase, aspartate aminotransferase or bilirubin. Full blood counts included haemoglobin and/or platelets. A patient was recorded as having a liver function test or full blood count if they had one or more of the constituent tests on the same day. Bone health profiles were identified when a patient had a phosphate test.

We divided the number of tests by the group size in each 3-month period (which declined over time) to calculate the mean number of tests per patient per quarter, with the group size in each period determined on the basis of the patient’s first and last visit dates, regardless of whether or not she/he had actually attended for care in that period. Unit costs were applied to each laboratory test and summed across all tests, and the mean cost per patient was calculated within each EIC and time stratum. The unit costs were mean costs from two HIV clinics in London, measured in 2016 GBP. The unit costs are summarised in Table 2.

Mean costs for the six pre-ART EIC groups were then plotted by quarter in the 4 years following ART start. Given the limitations of the data we do not undertake formal tests of differences between groups over time, but focus on descriptive trends. We were particularly interested in investigating whether or not pre-ART EIC affected testing and test costs after starting ART, and in particular whether or not costs became higher in the less engaged groups over time if their health declined, at a more rapid rate than in those who had a higher pre-ART EIC value.

Although we had planned to incorporate the development of a retention risk tool into the first phase of the study, the Management Team agreed that the REACH survey would provide a richer source of data than the UK CHIC data set to identify clinical and non-clinical factors that may predict disengagement. This analysis was therefore moved into phase 2 of the study and is described as follows.

Phase 2a: quantitative component – questionnaire and clinical data collection

Study questionnaire

The anonymised pen-and-paper questionnaire was completed in clinics where it was linked to clinical data. It contained 80 questions and took about 20–30 minutes to complete (see Appendix 2). Questions were based on variables from the COM-B (‘capability’, ‘opportunity’, ‘motivation’ and ‘behaviour’) model (Figure 2), which proposes that behaviour occurs as an interaction between three necessary conditions of capability, opportunity and motivation. 45 The COM-B model is at the centre of an integrative framework of behaviour change interventions, the behaviour change wheel comprising nine intervention functions and seven policy categories. It provides a framework for developing, evaluating and synthesising interventions, including selecting behaviour change techniques in the development process.

FIGURE 2.

The COM-B model of behaviour. Adapted from the original figure (see Michie et al. 45) through inclusion of description of the three domains. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

A framework for REACH primary research (see Appendix 3) was created to aid our variable selection. The framework brought together the COM-B model, the associated domains of behavioural influence from the theoretical domains framework46 and related constructs, as well as factors associated with engagement requiring change that were identified in an in-depth report on non-attendance of HIV services in Scotland. 37 Thus, when selecting which items should be included in the questionnaire, we could ensure that we had adequate measures from the different sections of the framework, including the three key domains of capability, opportunity and motivation.

Whenever possible, the questionnaire incorporated validated items used in other large-scale behavioural surveys but new questions were developed where we were unable to find published measures. Items from the following scales were included: three items from Household Food Insecurity Access Scale;47 all seven items from the Strive Internalised Stigma scale;48 three items from the environmental mastery subscale of the Psychological Well-Being Scales (PWB);49,50 five items from the Duke-UNC Functional Social Support Questionnaire (DUFSS);51 all four items from the Patient Health Questionnaire (PHQ4);52 all three items from the European AIDS Clinical Society screening questions for neurocognitive impairment;53 and two items from the Health Information Competence Scale. 54,55 We used the following five items from the 11-item Belief about Medicines Questionnaire (BMQ):56,57

1. my health, at present, depends on these medicines

2. having to take these medicines worries me

3. my health in the future will depend on these medicines

4. I sometimes worry about becoming too dependent on these medicines

5. these medicines give me unpleasant side effects.

We used all five items from the Medication Adherence Report Scale (MARS). 58 We did not use any of the framing statements from the BMQ or MARS due to the limited amount of space in the questionnaire.

The questionnaire content was piloted to test feasibility and acceptability among five PLWH, including a patient representative with extensive experience of working with people who have difficulties engaging in HIV care. The questionnaire was piloted to test keywords and constructs with five PLWH. The content was modified in the light of this feedback.

The final questionnaire was a 24-page printed A5 booklet. In addition, a computer-assisted self-interview (CASI) version was developed, programmed in survey software and tested for the study. However, most of the clinics preferred to use the paper version and the CASI version was only offered to a minority of patients at two of the clinics and electronic responses were returned by two patients. The questionnaire was divided into the following sections:

General information: gender, age, ethnic group, country of birth, immigration status, language, relationship status, children, pregnancy, work status, education, religion, sexual orientation, poverty, accommodation and hunger.

Life with HIV: date of diagnosis, place of diagnosis, activity, HIV disclosure, stigma, environmental mastery, caring responsibilities, domestic abuse, social support, support groups and recreational drug use.

Health and health care: health status, mental health, professional support, neurocognitive impairment, general practitioner (GP) registration, inpatient stays, health information competence.

Human immunodeficiency virus care: reasons for missed appointments, number of clinics attended, frequency of attendance, time at present clinic, travel to clinic, clinic hours, mode of consultations, clinic recommendation, appointment booking, reasons for attending, communication with reception staff, nurse and doctor, and treatment explanation.

Medicines: beliefs about medicines, ART status, home delivery service, medication adherence and reasons for not taking ART.

Phase 2b: qualitative component – individual patient interviews

Exploratory, face-to-face, semistructured interviews were undertaken with a purposively selected sample of men and women who had attended the HIV clinics where survey recruitment was taking place from June 2014 to February 2015. Eligible participants were men and women living with HIV aged ≥ 18 years. Patients were excluded from the individual interviews if they were unable to provide informed consent or if they had acquired HIV through vertical transmission as another study [Adolescents and Adults Living with Perinatal HIV (AALPHI) cohort study of HIV-infected young people]65 was recruiting these patients for individual interviews concurrently.

Phase 2c: community focus groups

The individual interviews were supplemented by community FGs. These groups were used to uncover alternative perceptions that may be articulated in a different, non-medicalised setting. Their stated purpose was to explore patients’ experiences, service preferences and perceived barriers to accessing HIV services.

For this part of the study, we considered recruiting up to four FGs of six to eight adults who did not attend the HIV clinic regularly or had stopped going to the clinic (now or in the past). We decided to recruit a total of three groups to represent the two main groups that are affected by HIV (gay men and black African men and women) and a third group to represent other people affected by HIV. These groups were to consist of:

• gay and bisexual men

• African men and women

• non-African men and women.

Key informant recruitment

Organisations and individuals within them were identified for each of the key constituencies. All prospective informants were approached by e-mail and, if they agreed, a 30-minute interview was arranged and conducted by telephone, Skype™ (Microsoft Corporation, Redmond, WA, USA) or face to face. All face-to-face interviews were conducted at mutually convenient locations. Participants were offered a copy of the interview transcript and a copy of the findings from the study. At the time of conducting the interviews, all participants were informed that their contribution would be anonymous. However, it became clear that some participants might prefer for their contribution to be acknowledged. They were therefore asked if they would prefer to be named in any publication of the findings and nine of the participants said that they would like to be identified.

Key informant content, processing and analysis

A topic guide was used to explore why patients miss appointments, what interventions or service improvements would help and what resources were needed (see Appendix 6). The interviews took about 30 minutes. Again, they were digitally recorded and transcribed verbatim.

As the key informant data were much shorter and less in-depth than the patient data – with participants talking about their professional experiences rather than in-depth and more lengthy discussion of their personal experiences – we used a more expedient method of analysis. Rather than coding all the transcripts in NVivo, we moved straight to summarising the data under theme headings that emerged from the data. AH conducted this analysis, reading and rereading the transcripts and organising the emerging data within the context of the COM-B model. In order to validate this analysis, FB and AE each read three transcripts (a total of six between them) and then examined the summary to see whether or not it fitted their understanding of the data.

Intervention costs

Using the methods described above, we put together four interventions that might be tested to improve engagement with care. We then undertook a preliminary evaluation of the costs of the four proposed interventions, after developing detailed descriptors of each. All costs were calculated according to 2015/16 GBP figures. Costs were calculated from a NHS and personal social services perspective. 68 The interventions would each incur a mixture of staff and non-staff costs. The former would include a range of staff types from consultants to unpaid peer support workers. The descriptors for each intervention included estimates of the amount of time each member of staff would spend on that intervention. To these we then applied unit costs per staff type obtained from published sources. 69 We valued the estimated non-staff costs (e.g. transport allowance, food allowance, telephone allowance, room hire, production of posters, leaflets and pocket guides) using market prices. We calculated the total cost of each intervention and then divided this by an estimate of the number of patients likely to receive the intervention to calculate the mean cost per patient. For three of the interventions, we calculated the costs over a 6-month period making assumptions about the number of contacts made by each patient during this time; for the fourth intervention, we calculated mean one-off costs per patient making assumptions about clinic size. The output is a series of indicative costs for each intervention, which can then be evaluated more formally in subsequent prospective studies.

Patient and public involvement

We have engaged with the public and patients at all stages of our project. Our patient and community engagement has been facilitated through the UK-CAB, a network for community HIV treatment advocates across the UK. Since the inception of the project, a representative from the UK-CAB has been involved as a co-applicant and member of our Management Team. She was responsible for recruiting PLWH to participate in our FGs and contributed to the design of the publicity material and the content of the FG topic guide. Our community representative and researcher/project manager jointly facilitated the FGs.

In addition, our Study Steering Committee and Advisory Group have both included community representatives who have contributed to the design and management of the research through these channels. They have provided valuable individual feedback our patient materials, including the content of the questionnaire and topic guides.

The questionnaire was also piloted on five PLWH and considerably revised on the basis of their input. We have interviewed three patient representatives who have contributed their expertise to the study as key informants.

At the time of writing, our findings have only recently been finalised but we will feed back to service user groups from the participating clinics over the coming months and lay summaries of our research findings will be available to patients attending for HIV care at participating sites. We will also host a dissemination and networking event for all key stakeholders in the coming months.

Background factors

The associations between background factors and attendance pattern are shown in Table 14. Just over a quarter of respondents were female (27.6%) and IAs and NAs were more likely than RAs to be female (IA 30.5% and NA 34.1% vs. RA 24.2%; p = 0.020). There was a significant association with age, with RAs more likely to be in the older age group (≥ 45 years, 53.3%) than IAs (40.9%) or NAs (37.8%) (p < 0.001) and IAs and NAs more likely to be younger. There was no significant difference between attendance groups when comparing white ethnicity, black African and other ethnicity, but RAs were more likely to be white than any other ethnicity (RA 57.2% vs. IA 50.0% and NA 47.2%; p = 0.032). NAs were more likely to have left full-time education by the age of 18 years (38.8%) than RAs (27.1%) or IAs (29.9%) (p = 0.021). There were no significant associations between attendance and sexual orientation, region of birth, year of arrival in the UK or religion.

In multivariable analysis, we used the continuous variable for age, with the adjusted odds ratio (aOR) relating to each 1-year increase in age. Adjusting for the clinic where respondents were recruited, indicated that increasing age was associated with decreasing likelihood of irregular attendance [aOR 0.96, 95% confidence interval (CI) 0.95 to 0.98; p < 0.001] and being female was associated with increased likelihood of irregular attendance (aOR = 1.94, 95% CI 1.32 to 2.87; p = 0.001). Increasing age was also associated with decreasing likelihood of non-attendance (aOR 0.96, 95% CI 0.95 to 0.98; p < 0.001), whereas being female (aOR 2.29, 95% CI 1.46 to 3.61; p < 0.001) and lower education (aOR 1.81, 95% CI 1.21 to 2.71; p = 0.004) were associated with increased likelihood of non-attendance.

Factors relating to HIV diagnosis and health service use

The associations between factors relating to HIV diagnosis and health service use are shown in Table 15. Three-quarters of the respondents had been diagnosed with HIV for > 5 years, with over half diagnosed for ≥ 10 years (53.2%) and a quarter (23.7%) diagnosed for 5–10 years. There was a significant association between time since diagnosis and attendance. IAs and NAs were more likely to have been diagnosed for ≥ 10 years than RAs (IA 59.5% and NA 55.5% vs. RA 49.5%; p = 0.002). One-quarter of women had been diagnosed with HIV during pregnancy (25.3%), but this was not significantly associated with attendance. Neither was place of diagnosis related to attendance, with about half of respondents diagnosed at a sexual health clinic (47.8%), 17.7% in hospital, 11.4% at a HIV testing service and 10.2% at the GP.

The majority of respondents had only ever attended one HIV clinic (69.9%), with two-thirds having attended their current clinic for at least 5 years, including 41.0% who had attended the same clinic for ≥ 10 years. There was no significant association between number of clinics attended or number of years at current clinic and attendance pattern. Respondents were asked how often they expected to have a routine consultation at their HIV clinic. Just over half the RAs expected 6-monthly visits (RA 53.7% vs. IA 38.6% and NA 45.6%), whereas half of the IAs expected visits every 3–4 months (IA 50.1% vs. RA 36.1% and NA 34.8%); NAs were the most likely to expect yearly visits (NA 12.7% vs. RA 6.6% and IA 2.7%; p < 0.001). Most respondents were either extremely likely (80.6%) or likely (15.3%) to recommend their clinic and this did not vary significantly between attendance patterns.

Although the majority of respondents were registered with a GP (95.5%), NAs were less likely to be registered (NA 89.0% vs. RA 96.7% and IA 97.0%; p < 0.001). Similarly, RAs and IAs were more likely to have attended their GP in the last year (RA 80.5% and IA 81.7%) than NAs (72.7%; p = 0.020).

We were interested in what brought the respondents into the clinic and we asked them about the circumstances and purpose of their last appointment. The majority had a booked appointment (89.7%) and reported that they did not feel unwell (84.7%). These circumstances were not significantly associated with attendance pattern. However, RAs and IAs were more likely than NAs to be attending a routine consultation at last visit than NAs (RA 90.0% and IA 88.1% vs. NA 78.7%; p = 0.003). Needing more HIV medication was reported as a very important reason for attending (73.0% of respondents) compared with wanting advice about HIV (64.6%), wanting other advice (38.7%) or needing practical support (21.4%). Reasons for attending did not vary significantly by attendance pattern except for medication requirement, with RAs and IAs more likely to need medication than NAs (RA 73.8% and IA 79.6% vs. NA 59.0%; p < 0.001).

The multivariable analysis, using the continuous variable for number of years since diagnosis and adjusting for the clinic, age, gender and education, indicated that increasing number of years since HIV diagnosis was associated with irregular attendance (aOR 1.05, 95% CI 1.02 to 1.07; p < 0.001). It indicated that increasing number of years since HIV diagnosis was associated with increasing likelihood of non-attendance (aOR 1.05, 95% CI 1.02 to 1.08; p = 0.001), and not being registered with a GP was associated with non-attendance (aOR 3.92, 95% CI 1.79 to 8.58; p = 0.001).

Factors relating to capability

In this section we look at factors relating to respondents’ capability of attending the clinic (Table 16). Only one-fifth of respondents did not speak English as their main language (20.1%) and this was not significantly associated with attendance pattern. Symptoms of neurocognitive impairment (memory loss, feeling mentally slower, difficulty paying attention) were reported by 41.3% of respondents and this was more likely among IAs (52.5%) than RAs (37.3%) or NAs (36.9%; p < 0.001). IAs were also less likely than IAs or NAs to report excellent or very good health in the past 4 weeks (IA 35.2% vs. RA 50.6% and NA 42.2%) and more likely to report fair or poor health (IA 38.3% vs. RA 24.5% and NA 24.2%; p < 0.001). More than two-thirds of respondents had ever had an inpatient stay (70.7%), one-fifth had been hospitalised in the past year (19.7%), nearly half reported that HIV affected their day-to-day activity to some extent (46.3%), and 14.4% reported a long-standing condition that caused difficulty with accessing buildings, streets or vehicles. None of these factors was significantly associated with attendance.

The majority of respondents had never injected drugs (87.0%) and this was not significantly associated with attendance pattern, but recreational drug use in the past 5 years was more common (37.6%) and more likely to be reported by IAs (41.7%) and NAs (46.2%) than by RAs (33.1%; p = 0.003).

We included two items measuring health information competence and three-quarters of respondents both understood where to get health information (76.6%) and felt in control of health information (75.4%), which did not vary significantly by attendance.

The multivariable analysis, adjusting for the clinic and background factors, indicated that respondents who reported symptoms of neurocognitive impairment were more likely to be IAs (aOR 1.79, 95% CI 1.26 to 2.55; p = 0.001), as were those who reported poorer health compared with very good or excellent health (aOR 1.89, 95% CI 1.25 to 2.85; p = 0.002). Non-attendance was associated with reporting recreational drug use in the past 5 years (aOR 2.03, 95% CI 1.27 to 3.25; p = 0.003) and weakly associated with reporting good health compared with very good or excellent health (aOR 1.57, 95% CI 98 to 2.50; p = 0.059).

Factors relating to motivation

Motivation describes the mechanisms that activate or inhibit a behaviour and, in this section, we explore how items relating to motivation are associated with attendance, as shown in Table 17. One-third of respondents said that religion was very important in their life (30.3%), but this was not significantly associated with attendance pattern. We asked patients if they had experienced any of a list of seven feelings in the past 12 months because of their HIV status. At least one of the seven items relating to internalised HIV stigma had been experienced by 43.8% of participants; the most likely to be reported was low self-esteem (34.1%), followed by self-blame (28.8%) and shame (28.7%); only 4.9% of respondents ticked, ‘I feel I should be punished’. IAs were more likely than RAs or NAs to report low self-esteem (IA 40.0% vs. RA 32.6% and NA 29.4%; p = 0.046) and feeling suicidal (IA 16.5% vs. RA 9.4% and NA 11.9%; p = 0.013), but there was no significant difference between the attendance patterns on an overall internalised stigma summative score.

We asked respondents about symptoms of anxiety and depression using the PHQ4 scale. 52 None of our respondents reported severe levels of anxiety and depression, although it would not have been appropriate for clinics to approach patients with severe symptoms about participation. There was an indication that IAs were less likely than RAs or NAs to report normal levels (IA 70.1% vs. RA 79.5% and NA 79.1%; p = 0.057); they were also more likely to report that they had ever been diagnosed with depression (IA 38.2% vs. RA 29.1% and NA 31.4%; p = 0.043).

We included three items from a scale measuring environmental mastery in our questionnaire. Although findings from this a subscale of the PWB49,50 are reported here under motivation, environmental mastery is also likely to affect capability. IAs scored lower on a summative score of environmental mastery than RAs or NAs (IA 10.1 vs. RA 10.7 and NA 10.3; p = 0.027); they were more likely to report that the demands of life often got them down (IA 46.2% vs. RA 36.1% and NA 42.7%) and less likely to feel in charge of the situation in which they lived (RA 58.6% vs. RA 66.7% and NA 61.8%, p = 0.027).

In multivariable analysis, irregular attendance was associated with ever being diagnosed with depression (aOR 1.61, 95% CI 1.11 to 2.33; p = 0.012) and with being uncertain rather than agreeing about being in charge of life (aOR 1.75, 95% CI 1.13 to 2.72; p = 0.013).

Factors relating to opportunity: social influences

Social influences shape an individual’s opportunity to enact behaviours. In this section, we explore respondents’ social environment in terms of the people they interact with both within and outside the HIV clinic, as shown in Table 18.

Half of the respondents were not currently in a relationship (51.3%), whereas 31.0% had been in a relationship for at least 5 years. Although this was not significantly associated with attendance pattern, RAs were more likely to be cohabiting (37.6%) than IAs (30.2%) or NAs (29.2%; p = 0.041). Among respondents with a partner, almost half had a partner who was also HIV positive (45.2%) and most had told their partners about their HIV status (88.7%), neither of which varied significantly by attendance pattern. One-fifth of respondents had been abused by, or afraid of, a partner or ex-partner in the past year (20.6%) but this was not significantly associated with attendance.

Although most respondents had told someone about their HIV status (86.9%), 42.0% had not told any family members and 30.0% had told none of their friends. RAs were more likely to have told none of their family members about their HIV status (RA 45.8% vs. IA 37.1% and NA 36.8%; p = 0.035) but there was no significant difference between groups on telling friends about their HIV status. Most respondents had told their GP (89.3%) and this did not vary significantly by attendance.

We included five items from the DUFSS. 51 Across the attendance patterns, the majority of respondents reported that they had enough people who cared about what happened to them (73.2%). Somewhat fewer felt that they got enough love and affection (66.7%), enough invitations to go out (66.3%) and enough chances to talk (64.1%). IAs and NAs were less likely than RAs to report getting enough help when they were sick in bed (IA 58.2% and NA 58.4% vs. RA 66.3%; p = 0.046), although there was no significant difference between the attendance groups on an overall score for level of social support. Two-thirds of respondents overall had received information from their clinic about HIV support groups (67.9%), but only 30.9% had used a support group in their local area and 16.5% had not used a group but would like to. Among those who had used a group, 53.1% definitely found it helpful compared with 11.6% who did not find it helpful. None of these variables was significantly associated with attendance pattern.

We asked respondents a range of questions about their interactions with clinic staff on the last occasion that they came to the clinic. Most reported that the reception staff were definitely friendly and approachable (82.1%), that they completely understood explanations for any treatment or action (90.1%), that the doctor was definitely friendly and approachable (95.3%), the doctor definitely listened to them (94.0%) and that they definitely had confidence and trust in the doctor (91.9%). This did not vary significantly between attendance groups. All respondents were similarly likely to report that they had confidence and trust in the nurse (90.7%), but NAs were less likely than RAs or IAs to report that the nurse was friendly (NA 89.5% vs. RA 94.9% and NA 94.6%; p = 0.037) or listened to them (NA 86.4% vs. RA 93.2% and IA 94.6%; p = 0.006).

In multivariable analysis, irregular attendance was associated with having told family members about their HIV status (aOR 1.52, 95% CI 1.06 to 2.20; p = 0.025) and with not having enough help when sick in bed (aOR 1.64, 95% CI 1.14 to 2.37; p = 0.008). Non-attendance was associated with having told family members about their HIV status (aOR 1.63, 95% CI 1.02 to 2.60; p = 0.043), not having enough help when sick in bed (aOR 1.76, 95% CI 1.12 to 2.79; p = 0.015) and not being listened to by the nurse (aOR 2.17, 95% CI 1.06 to 4.42; p = 0.034).

Factors relating to opportunity: barriers

In the same way that social influences shape an individual’s behaviour, structural factors can also enable or hinder attendance at the HIV clinic. In this section, we explore these barriers and facilitators of HIV care, as shown in Table 19.

Regular attenders were less likely to have children (25.4%) than IAs (31.8%) or NAs (37.7%; p = 0.003). However, there was no significant difference between attendance groups on responsibility for looking after children, family members or others that was reported by 24.9% of respondents. Two-thirds of respondents were British citizens (67.3%), 15.2% were European Union citizens, 11.1% had indefinite leave to remain and 6.5% had temporary or no official immigration status. This was not significantly associated with attendance pattern. The majority of respondents were working (58.9%) and there was some indication that RAs were less likely than IAs or NAs to be students (RA 3.4% vs. IA 5.8% and NA 7.5%) and less likely to be unemployed (RA 19.4% vs. IA 25.8% and NA 23.1%; p = 0.056).

Several indicators of economic status were included in the questionnaire, which showed that respondents were generally financially disadvantaged but that this was more likely among IAs and NAs. Less than half of the respondents had enough money for their basic needs all the time (43.1%) and this was less likely among IA (35.3%) and NAs (34.2%) than among RAs (49.5%; p < 0.001). Moderate or severe hunger was reported by 13.6% of RAs, compared with 28.5% of IAs and 23.8% of NAs (p < 0.001). One-third of RAs (33.3%) were homeowners, compared with 22.1% of IAs and 19.1% of NAs (p < 0.001). There were a lot of missing postcode data to create and Index of Multiple Deprivation score for respondents’ neighbourhoods, but 43.5% lived in a neighbourhood that was among the most deprived in England. Time in accommodation was not significantly associated with attendance group and 51.2% had lived in their current accommodation for > 5 years.

We asked respondents several questions relating to the convenience of the clinical service provision. RAs were less likely than IAs and NAs to travel to the clinic by public transport (RA 75.2% vs. IA 82.0% and NA 86.0%) and were more likely to go on foot or by bicycle (RA 12.6% vs. IA 5.7% and NA 6.7%), whereas NAs were less likely to use a car or motorbike than RAs or IAs (NA 6.7% vs. RA 11.1% and IA 11.5%; p = 0.022). There were no significant differences by attendance on the time it took to get to the clinic or convenience of the opening hours. The majority took < 1 hour to get to the clinic (81.7%), with 38.9% taking < 30 minutes, but 15.6% took 1–2 hours and 2.7% took > 2 hours. Most respondents found the opening hours very or fairly convenient (95.0%). A minority of all respondents wanted Sunday opening (15.0%) and somewhat more wanted weekday early morning opening (22.8%). One-third of all respondents wanted Saturday opening (34.0%) and mid-week evening opening (33.5%), with the latter more popular among IAs and NAs than RAs (IA 38.3% and NA 37.2% vs. RA 30.0%; p = 0.033).

In terms of previous and desired mode of communication for consultations, the only significant difference was that NAs were less likely to have used e-mail than RAs or IAs for this purpose (NA 12.3% vs. RA 19.1% and IA 22.1%; p = 0.049). Comparison of previous and desired mode of communication indicates that 93.6% of respondents have had face-to-face consultations and 85.0% would like face-to-face consultations in the future; 37.0% have had telephone consultations and 48.7% would like telephone consultations; 16.1% have had consultations by text and 26.1% would like this in the future; 18.8% have had e-mail consultations and 39.4% would like consultations by e-mail in the future; and only three respondents reported consultations by Skype or similar and 23.2% would like this type of consultation in the future.

Three-quarters of RAs booked their last appointment in person (75.7%), compared with 63.5% of IAs and 49.1% of NAs, whereas NAs were more likely than RAs and IAs to book by telephone (NA 46.5% vs. RA 23.4% and IA 36.1%) or by e-mail/text (NA 4.4% vs. RA 0.9% and IA 0.4%; p < 0.001). However, there were no significant differences between attendance groups on ease of booking, which was mostly reported as being very easy (72.5%).

In multivariable analysis, irregular attendance was associated with having children (aOR 2.11, 95% CI 1.39 to 3.20; p < 0.001) and not having enough money for basic needs (aOR 1.74, 95% CI 1.19 to 2.56; p = 0.005). Non-attendance was associated with having children (aOR 2.76, 95% CI 1.69 to 4.50; p < 0.001), not having enough money for basic needs (aOR 1.63, 95% CI 1.02 to 2.61; p = 0.042) and not owning one’s own home (aOR 1.87, 95% CI 1.09 to 3.20; p = 0.023).

Use of and beliefs about antiretroviral therapy

Table 20 shows the respondents’ use of, and beliefs about, ART. There were no significant differences between attendance groups on their scores relating to their belief about medicines, either for the general benefit subscale or for the general distrust subscale. RAs were more likely to be taking ART than IAs or NAs (RA 94.2% vs. IA 90.7% and NA 64.6%), whereas NAs were more likely than RAs or IAs to have taken ART in the past (NA 15.9% vs. RA 0.9% and IA 4.5%) or not at all (NA 19.5% vs. RA 4.9% and IA 4.8%; p < 0.001). Multivariable analysis, adjusting for the clinic, age, gender and education, indicated that IAs were more likely to have been on ART previously than those who were currently on ART (aOR 3.83, 95% CI 1.29 to 11.40; p = 0.016), but they were not more likely to have never been on ART. NAs were more likely to have been on ART previously (aOR 18.17, 95% CI 6.44 to 51.28; p < 0.001) and were more likely to have never been on ART (aOR 5.35, 95% CI 2.89 to 9.90; p < 0.001).

Table 21 shows that, among those on ART, IAs and NAs were more likely than RAs to have been taking it for longer: 5–10 years (IA 31.7% and NA 32.0% vs. RA 24.6%) or > 10 years (IA 39.4% and NA 45.9% vs. RA 35.9%; p = 0.005). RAs on ART were more likely to be using the home delivery service (59.5%) than IAs (46.6%) or NAs (38.7%) on ART (p < 0.001). IAs and NAs reported poorer adherence to ART with lower scores on the MARS than RAs (IA 22.6 and NA 22.3 vs. RA 24.0; p < 0.001) and greater likelihood of missing a dose on 1 or more days in the past week (IA 35.5% and NA 38.5% vs. RA 12.8%; p < 0.001). They reported lower scores on the ART necessity subscale (IA 17.4 and NA 17.2 vs. RA 18.0; p = 0.040) and higher scores on the ART concerns subscale (IA 10.0 and NA 9.8 vs. RA 9.1; p = 0.005). 56,57

Among the minority of respondents (n = 102) who were not taking ART and responded to the follow-up questions, 46.1% said that they had been advised to take ART. NAs made up the largest proportion of respondents who were not taking ART against the doctor’s advice (NA 60.0% vs. RA 13.3% and IA 26.7%). Table 22 shows the reasons that these respondents gave for not taking ART against advice at this time. As the numbers are small, they are not divided up by attendance group. The reason that was most likely to be reported as very important was the side effects of medication (61.7%), followed by feeling depressed or overwhelmed (54.3%), not wanting to think about being HIV positive (32.6%) and not wanting anyone to see the medication (29.8%). Reasons that were very important for a smaller proportion of respondents were homelessness (11.4%), harmfulness of medication (9.1%), use of alcohol or drugs (8.9%), information from the media (8.7%), information from friends (8.5%), complication of taking medication (6.8%) and, finally, use of alternative therapy (2.3%).

Clinical data

In addition to fielding our questionnaire, we collected clinical data from patients’ notes. The item above on whether or not patients were currently taking ART is an amalgamation of patient-reported and clinical data, and the item on years since starting ART is from the clinical data set. Both items are included in the above description of the data, as they fit into the section on use of and beliefs about ART. All other data taken from patients’ notes are reported below and shown in Table 23.

Although the majority of patients (80.5%) did not meet any of the HARS 3 categories for being classified as complex patients, almost one in five did. RAs were less likely than IAs or NAs to be reported to meet any of the HARS 3 complex patient criteria (RA 16.0% vs. IA 24.5% and NA 23.2%; p = 0.007). NAs were more likely than RAs or IAs to be pregnant (NA 5.5% vs. RA 0.5% and IA 0.4%; p < 0.001), suggesting that pregnancy had brought them back into care; and IAs and NAs were more likely than RAs to have persistent viraemia after 6 months on ART (IA 8.6% and NA 8.5% vs. RA 4.0%; p = 0.012). IAs were more likely than RAs or NAs to have ever had a HIV-related inpatient stay (IA 30.9% vs. RA 21.4% and NA 26.3%; p = 0.022).

There were no significant differences between attendance groups in CD4 count at diagnosis or the first recorded CD4 count (where the CD4 count at diagnosis was not available), or on CD4 count at ART initiation. One-third of patients had a CD4 count of < 200 cells/mm³ at diagnosis (33.0%), one-fifth had a first recorded CD4 count of < 200 cells/mm³ (19.2%), and 39.9% had a CD4 count of < 200 cells/mm³ at ART initiation. RAs and IAs were more likely to have a current CD4 count of > 350 cells/mm³ than NAs (RA 88.6% and IA 84.7% vs. NA 67.3%; p < 0.001). NAs were less likely than RAs or IAs to have a viral load of > 5.0 log-copies/ml at diagnosis (NA 16.9% vs. RA 42.5% and IA 40.8%; p < 0.001). There was no significant difference between groups for first recorded viral load. RAs were more likely to have a current viral load that was undetectable (RA 86.2% vs. IA 76.9% vs. NA 45.1%; p < 0.001).

About one-fifth of patients had ever had an AIDS-defining illness (22.9%) and this did not vary significantly according to attendance group. The most commonly reported AIDS-defining illness was Pneumocystis carinii pneumonia in 6.6% of patients, followed by pulmonary tuberculosis in 4.6%, Kaposi’s sarcoma in 3.1%, oesophageal candidiasis in 2.7% and herpes simplex virus in 2.0%, none of which varied significantly by attendance pattern. A minority of patients had hepatitis C current coinfection (5.8%) or hepatitis B coinfection (3.6%), but neither was significantly associated with attendance. One-fifth of patients had experienced mental health issues in the past year (20.5%) and this was not statistically associated with attendance. Whereas 5.9% of RAs had issues of drug or alcohol dependency in the past year, this was true of 16.7% of IAs and 18.5% of NAs (p < 0.001).

In multivariable analysis, irregular attendance was associated with being a complex patient according to the HARS 3 criteria (aOR 1.76, 95% CI 1.16 to 2.67; p = 0.008), having a HIV-related inpatient stay (aOR 1.51, 95% CI 1.01 to 2.27; p = 0.045) and drug/alcohol dependency issues (aOR 2.90, 95% CI 1.67 to 5.05; p < 0.001). Non-attendance was only significantly associated with drug/alcohol dependency issues (aOR 3.95, 95% CI 2.17 to 7.21; p < 0.001).

Individual patient interviews

We undertook a total of 33 interviews with PLWH. They were made up of 10 RAs, 13 IAs and 10 NAs. We had originally planned to recruit up to 20 IAs (making an overall total of 40 interviews), but after recruiting 13 IAs, we were hearing similar stories from the participants and decided that we had reached sample saturation. Our NAs included people who had ever experienced a 1-year period of non-attendance in comparison with our survey participants who had experienced a recent period of non-attendance. When we began interviewing people, we found that they were able to talk in detail about their experiences of non-attendance, even if it had occurred several years before.

In addition to attendance pattern, our other primary sampling criterion was combined gender and sexual orientation. We aimed to recruit roughly equal proportions of gay or bisexual men and heterosexual men and women. However, women were more willing to participate and our sample consists of nine gay men, eight heterosexual men and 16 heterosexual women. No men who identified as bisexual took part in the interviews and, with regard to ethnicity or country of birth and age, the subsamples were made up as follows:

Gay men:

• seven UK born; two non-UK born

• two aged ≤ 35 years; seven aged > 35 years.

Heterosexual men:

• four black African; four other ethnicity

• three aged ≤ 35 years; five aged > 35 years.

Women:

• 10 black African; six other ethnicity

• five aged ≤ 35 years; 11 aged > 35 years.

Among these participants, we aimed to recruit between three and six people who had ever injected drugs, between three and six patients who were not on ART and between three and six who were diagnosed within the past 2 years. We succeeded in recruiting three people who had injected drugs, two patients who were not on ART (a long-term non-progressor and a woman who had been diagnosed 3 years before during pregnancy) and four who had been diagnosed in the past 2 years.

Participants had been diagnosed for a median of 7 years (ranging from 4 months to 30 years). Their median age was 40 years (ranging from 25 to 56 years). Almost half were of white ethnicity (45.5%), 39.4% were black African, 9.1% were black Caribbean and 6.1% were of mixed ethnicity. The majority were born outside the UK (72.7%). Over half were not in a relationship (57.6%), 21.2% were in a relationship but not living with a partner and 21.1% were living with a partner. Two-thirds had some further or higher education (66.7%) and the majority were not working (72.7%).

Our participants are reasonably representative of an ageing HIV positive population. We did not manage to recruit as many young people as we would have liked, but this may be partly explained because we excluded patients who had acquired HIV through vertical transmission as the AALPHI study was recruiting these patients at the same time. It should also be noted that almost three-quarters of our participants were not working. This is not surprising, given that those who are unemployed are more likely to have the time to take part in an interview, but it means that our qualitative data include fewer perspectives from working people who may also have difficulties with attendance for this reason.

Focus groups

We conducted two FGs – the first (FG1) consisted of four gay men and the second (FG2) of six black African men and women (including two men and four women). Although we encouraged discussion among FG participants, most of their comments were addressed directly to the FG facilitators and the data were very similar in format to the interview data. The data were therefore incorporated into the interview analysis, although some different stories may have been elicited by recruiting these participants through community outreach rather than at their HIV clinics. Participants had been diagnosed for a median of 14 years (ranging from 6 to 27 years). Their median age was 46 years (ranging from 35 to 55 years). All had experienced periods of a year or more when they had not seen a doctor about their HIV and had missed and not rescheduled at least one appointment at the HIV clinic in the past year.

The names used below for both individual interview and FG participants are pseudonyms. In order to provide more context for the data, we will describe whether they were interviewed for the study as a RA, IA, NA or FG participant (FG1 or FG2); and include the following shorthand to describe them, followed by their age (no men who identified as gay or bisexual black Africans took part in this part of the study):

• female, black African: Fem-BA

• female, other ethnic group: Fem-Other

• heterosexual male, black African: HetMale-BA

• heterosexual male, other ethnic group: HetMale-Other

• gay male, UK born: GayMale-UK

• gay male, non-UK born: GayMale-NonUK.

Key informant interviews

We conducted a total of 19 semistructured interviews with service providers and funders. The key constituency of ‘voluntary sector’ was revised to ‘community support’ to reflect the fact that some patient reps are employed by the NHS to work in HIV clinics. The key informant representation from across the key constituencies in the field of HIV service provision was as follows:

Clinical:

1. clinical nurse specialist, NHS

2. senior sexual health adviser (Ceri Evans), Chelsea and Westminster Hospital NHS Foundation Trust

3. HIV nurse specialist, NHS

4. HIV consultant, NHS

5. receptionist, NHS

6. consultant in genitourinary medicine (GUM) and HIV (Vanessa Apea), Barts Health NHS Trust.

Public health:

1. public health expert 1, public sector

2. public health expert 2 (Gillian Holdsworth), SH:24.

Academia (health service researchers):

1. Professor of Medicine (Michael Mugavero), University of Alabama at Birmingham, USA

2. clinical research fellow/Consultant in HIV (Shema Tariq), University College London

3. academic research nurse, university.

Community support/health promotion:

1. patient representative 1

2. patient representative 2

3. chief executive (Juliet Reid), Centre for All Families Positive Health

4. health services manager (Mesfin Ali), Embrace UK

5. patient representative 3 (Chris Sandford), Mortimer Market Centre

6. peer caseworker (Sophie Strachan), Positively UK.

Policy:

1. director of policy, non-governmental organisation

2. NHS manager, NHS.

Key informants are identified by their job title. Their initials are also included if they have agreed to be identified. In the following description of the findings, we refer to ‘key informants’ who took part in this part and we refer to ‘patients’ who took part in the individual interviews and FGs. As previously described, all the people who took part in the individual interviews and FGs were currently engaged in care, and the word ‘patients’ will therefore be used to distinguish their contribution to the study from that of the ‘key informants’.

Physical capability

Psychological capability

Motivation

Opportunity: social influences

Opportunity: barriers

Although specific barriers may affect only small numbers of the patients who took part in our study, they could have a dramatic effect on attendance and were often associated with socioeconomic status. The barriers to attending were frequently cited by key informants as reasons for disengaging from care. Many key informants described how attending the clinic was not always a priority when people were struggling with the pressures of daily life and juggling priorities – their basic needs took precedence:

I think there are structural factors that if folks don’t have a safe place to live, a stable source of transportation, and a stable source of food, that a lot of times those structural things can take precedence over coming to the clinic.

Professor of Medicine (MM)

I think in terms of your hierarchy of needs, coming to HIV appointments falls lower and lower when you’re actually sort of scrabbling to survive.

Clinical Research Fellow/Consultant in HIV (ST)

Sometimes an escalation of difficulties or crisis could prompt disengagement.

Motivation

Although not everyone articulated a direct cause and effect, there was a clear and pervasive relationship between the stigma associated with being HIV positive and disengagement from care. This self-stigma was associated with negative thoughts such as not being able to tell anybody about being HIV positive and not being able to live the life one wanted to lead. Stigma affected both those who engaged regularly with care and those who did not. Stigma was a barrier to people accessing support when it was more comfortable not to think about HIV or when they worried about been seen at the HIV clinic. Patients described how the associated depression and low self-esteem could make taking care of oneself seem pointless. Although patients appeared to be particularly likely to engage with this type of thinking and dropping out of care when they were newly diagnosed, it could continue to affect engagement for people like Julie who had been diagnosed for 4 years:

I just think, why should I bother because even though I go there, this thing what I have, is not going to disappear. It’s not going to disappear. I’ve still got it, so I don’t know. I don’t . . . I just, sometimes I don’t, like, I don’t put it in my head that I’ve got HIV.

Julie (NA, Fem-BA, 25 years)

The possibility of passing on HIV to a child was also found to be difficult to deal with. For example, Jade (FG2, Fem-BA, 37 years) did not attend for 18 months after she had her second baby, because she had found the frequent HIV testing and possible HIV diagnosis of her first baby very stressful.

These feelings of self-stigma were especially strong among the black African patients and could be exacerbated by stigma experienced or imagined from other community members. Adele (NA, Fem-BA, 27 years) felt threatened by a member of the hospital staff from her home country that she confided in when she was having her baby and subsequently withdrew from care for 2 years.

Self-stigma was also found to be implicated with not attending the clinic when patients felt well, particularly before starting ART, even though they knew that they were expected to attend the HIV clinic for routine check-ups:

I was down, I was still in the process of denial . . . so I completely thought to myself, you know, what’s the point? I just need to, sort of like, stay strong for the kids as well as myself, and being that I kept well, I’ve never felt unwell and all that. So I kept thinking, oh, I’m fine. I’ll go next month. I’ll go a month after, and month turned into months turned into . . . and yes.

Josephine (NA, Fem-BA, 42 years)

However, Charles (NA, Male-BA, 47 years) understood that it was not necessary to attend because he was well and only attended twice in the 4 years after his diagnosis. Similarly, Miriam explains how the message she took away when she did not start ART at diagnosis was that she did not need to engage with care:

I just completely forgot, not forgot, just said, OK fine, I’m OK, because if they think if I’m not ready for medication it means I’m OK, you know, it means I’m fine.

Miriam (NA, Fem-BA, 39 years)

In relation to this, Rick (NA, GayMale-UK, 44 years) said that 6-monthly appointments were the ‘worst thing that could have happened to me’ because it seemed like he was doing well and could stop taking his medication.

Once ART was prescribed, patients’ experiences illustrate how their motivation to attend could be shaped by ART. Cynthia (FG2, Fem-BA, 35 years) stopped attending for a year when she was advised to start taking ART and was fearful of doing so. Among people who had started ART, the stress of taking it could result in a conscious decision to stop taking it and stop attending:

Along the way I think emotionally I got fed up with the drugs and stopped taking it without telling them.

Joyce (NA, Fem-BA, 36 years)

Jade (FG2, Fem-BA, 37 years) also stopped taking ART because it was ‘permanent persecution every day’, reminding her about HIV. Difficulties with ART could result in more gradual changes to adherence that could equally result in disengaging from care:

So I kind of drifted off that. I stopped going into the clinic because I didn’t want to admit to them that I drifted off that.

Rick (NA, GayMale-UK, 44 years)

As Rick describes, the relationship between non-adherence and non-attendance may be mediated by the desire to avoid the perceived censure of health-care professionals when not taking medication. The hierarchical relationship between doctor and patient was apparent across the diverse cultural backgrounds of the patients, with references to being a good patient who attends regularly and takes their medication and not wanting to be told off for not doing so. The association between adherence, attendance and the doctor–patient relationship also illustrates the interaction between the elements of the COM-B model.

Opportunity: social influences

Although patients generally thought well of clinic staff, two of them described how periods of disengagement resulted from a breakdown in their relationship with health-care professionals at their HIV clinic. Both felt that they were not listened to and therefore withdrew from care.

Opportunity: barriers

Various barriers were implicated in stopping attending that were often particular to individuals but all resulted in complete disengagement from care. Charles (NA, HetMale-BA, 47 years) became homeless, which made it difficult to keep medication that required refrigeration and to keep up attendance. Although Joyce (NA, Fem-BA, 36 years) made a conscious decision to stop attending when she stopped taking her ART, this was preceded by changes in her child care arrangements which impacted her working hours. Some patients, such as Rick (NA, GayMale-UK, 44 years), describe a range of causes, including self-esteem, adherence and working in a responsible job with erratic hours, which made it difficult to adhere to complicated medical regimes and attend for HIV care. Barriers to attendance are interwoven with motivational issues, social influences and not prioritising HIV care. For example, Josephine (NA, Fem-BA, 42 years) was diagnosed during pregnancy – she was doing two jobs, her husband was not working and abusive, she had four children including the baby and one with a serious disorder that needed a lot of care:

So whenever I had a free time, I was too exhausted even to think of coming to the clinic . . . I was always tired and the thing overwhelmed me. My mind was just concentrating at the immediate basic things in the house. My children, mortgage, work.

Josephine (NA, Fem-BA, 42 years)

Background

The evidence from our survey complemented the evidence from secondary analysis of UK CHIC data on the background factors associated with regular clinic attendance. In both cases, the analysis suggested that the association with gender could be explained by other factors, whereas there was a clear indication that younger patients were more likely to disengage from care. On the other hand, survey respondents were also more likely to disengage the longer they had been diagnosed whereas higher engagement among UK CHIC participants was found only in the first year after being initially seen at a clinic. Both analyses found an association between acquisition of HIV through sex between men and being on ART with better EIC. The survey found better engagement among white participants and UK CHIC found that those of ‘other ethnicity’ had poorer engagement. In addition, the survey indicated that lower education and not being registered with a GP were associated with a period of disengagement. The data suggest that interventions might be usefully targeted at these patient groups.

Physical capability

The findings from the survey were supported and enhanced by the findings from the qualitative component of the study. The data indicated that, when patients felt unwell, they are more likely to miss appointments, but ill health could also be the trigger for people to re-engage with care. Poorer health was reported among IAs and feeling unwell was given as a reason for not attending appointments by survey participants and those taking part in qualitative research. The data from clinic notes also showed that irregular attendance was more likely among patients who had had stayed in hospital because of their HIV. The study suggests that patients may need extra help – both psychological and practical – to attend the HIV service when they are not well.

Psychological capability

The quantitative and qualitative data indicated that forgetting to attend was a key reason for missed appointments. The survey data furthermore suggested that this may be associated with neurocognitive impairment. The evidence indicates that appointment reminders could be usefully employed to reduce missed appointments.

Data from the questionnaire and clinic notes indicated that alcohol and recreational drugs use were associated with disengagement from care. Although only one interviewee discussed recreational drug use, he gave this as a key reason for missed appointments, and interventions targeting drug and alcohol use should be considered.

Patients who took part in interviews and FGs showed the importance of information in tackling self-stigma and becoming empowered to take care of one’s health. Education can provide patients with the knowledge to engage with care.

Motivation

Depression emerged as a key reason for irregular attendance from both the quantitative and qualitative data sets. Although depression is more common among PLWH,70 the data from REACH suggested that people who disengage from care may be less well equipped to manage this and take charge of their health care. Mental health services have a key role to play in maintaining EIC.

Interview and FG participants expressed a clear association between not adhering to ART and not attending HIV services. Interventions tackling poor adherence to ART might help some people engage with HIV care more generally, as well as interventions targeting the language used by clinicians when addressing non-adherence with their patients.

Narratives of stigma were widespread among those participating in the qualitative research and self-stigma could contribute to disengagement from care. Items from the quantitative measure of internalised stigma indicated high prevalence of low self-esteem (34.1%), self-blame (28.8%) and shame (28.7%), but the summative score did not distinguish between groups. NAs were, however, more likely to give issues relating to stigma, such as not wanting to think about HIV or not wanting to be seen at the clinic, as reasons for not attending. This suggests that interventions that deal with stigma could help people to engage in HIV care.

Opportunity: social influences

Although many patients felt that the dedicated HIV clinic created a safe space, as described above, fear of involuntary disclosure of HIV status could stop people attending the clinic. Holding clinics in premises not associated with HIV might encourage some people to attend. Although there was little in the survey about peer support, its potential to address stigma in addition to other important benefits was consistently highlighted in the qualitative component of the study.

The quantitative and qualitative data generally suggested that HIV staff created a positive environment for patients. However, the relationship with health-care professionals could break down and interventions targeting staff interactions with patients might help some patients attend more regularly.

Opportunity: barriers

The quantitative and qualitative research emphasised the negative impact of poverty on EIC. Patients dealing with social issues like homelessness or immigration problems did not always prioritise their health care. Although we cannot tackle the overall problem of economic inequality, free transport to the clinic might reduce missed appointments. In addition, multidisciplinary, holistic support including social workers, psychologists and peer workers might provide advocacy and support to help patients engage in care.

Other barriers, including lack of child care and work responsibilities, were raised in both the survey and interviews. More flexible services, which offer services outside usual hours or in different locations, might help to address these obstacles.

Suggestions that address capability

Suggestions that address motivation

Suggestions that address opportunity: social influences

Suggestions that address opportunity: barriers

Identifying non-attenders

Key informants raised the issue of pre-empting non-attendance by identifying those at risk of disengaging and keeping an eye on them, as well as considering that risk may change over time. They highlighted the importance of taking a systematic approach to tracing people who had disengaged from care by drawing up an explicit, consistent follow-up policy involving a service with accountable staff who had time dedicated to this work. A service manager with clinical support or a lead clinician with multidisciplinary support could lead this.

The key informants talked about the sort of systems that might work and who should be involved. For example, there could be an electronic solution whereby a member of staff was responsible for an active process of flagging and tracking NAs. This might require integrating multiple data systems and improving data sharing between clinics. Technology might also be used to help keep patients contact details up to date, asking them to text their new address or telephone number.

Regular clinics or meetings could be held to identify NAs who should then be contacted by telephone, e-mail and/or letter. All attempts to contact patients should be documented. There were different suggestions about who should make the contact with patients. There was some benefit in the consultant who knew the patient making the approach, whereas in other services a nurse took responsibility for tracking people down. Patient representatives could also take on the role or health advisors had the skills to make an initial approach and then make an appointment to see the patient when they came into clinic. GPs could also be involved in tracing patients.

Redistribution of resources

Another central topic of discussion was the redistribution of resources. Reconfiguring services to reduce the cost of stable patients might free resources to be available for the most vulnerable patients who had engagement issues.

Interventions that address capability

One of the factors associated with disengagement is forgetting. This is one of the possible outcomes of substance abuse that could be tackled at the clinic by, for example, introducing a drug and alcohol worker into the clinic.

Although text messages can remind patients of their appointments, a personal reminder might be more powerful. An intervention was described whereby patients received a personal reminder call from someone that they knew in advance of their appointment and a personal call from the same person within 24–48 hours of missing an appointment. Others reported that an informal personal reminder has been found to be effective:

So some people tell me personally, oh, can you remind when it’s getting to, so maybe, a next day or a week to the appointment. I just give them a call and said, oh, you have appointment, so you need to come. And they will come.

Receptionist

The qualitative data described earlier indicated that knowledge could empower patients to take control of their health and engage with care. Peer support was one of the key means of imparting knowledge and we will come on to look at that under social influences. Areas of learning which might help patients to engage with the HIV service included education about the implications of disengagement, about going to clinic even if you are well, about the confidentiality of the clinic and around self-stigma.

Reinforcing an expectation of attendance might also encourage engagement. Professor of Medicine (MM) described creating a clinic culture that reinforced this expectation through the use of posters and staff messages and public health expert 1 talked about a proactive telephone call on one missed appointment to suggest that attending is the norm, with staff reinforcing this message.

Interventions that address motivation

Various ways of addressing stigma were suggested. These included raising public awareness through national campaigns or role models with the aim of normalising HIV and showing how it is now a treatable condition. The Time to Change campaign against mental health discrimination was given as an example. Media campaigns could also be used to engage different groups. Changes might also be made at the local level to destigmatise the clinic and to reassure patients that they are safe when they come to clinic.

Awareness should be raised among other health-care providers who needed updating on HIV. They could sit in at HIV clinics or attend specially designed workshops. Teaching to medical students could be reconfigured to include HIV stigma.

One key informant talked about providing HIV services in a way that was more similar to other long-term conditions might also have the effect of ‘normalising’ HIV by demything the condition and making care more accessible, although it may still need special provision:

There’s always this battle in one’s own head about shall we normalise HIV and, you know, do we need all this kind of special treatment? And then, every week you come across appalling instances of how people with HIV are treated, it makes you realise it sadly isn’t normal yet.

Patient representative 1

Appropriate and improved mental health provision was widely recognised as being needed for PLWH, reflecting the impact on mental health from moment of diagnosis and an elevated need for psychology in the population as a whole. Although mental health professionals were the key providers, additional support from a member of staff who was trained in psychological well-being could provide back-up to this service.

Self-efficacy was also seen as an important element of motivation, incorporating the idea of patients as their own caseworker. Clinic staff could be trained to raise health literacy and promote self-management. Specific interventions aimed at empowering patients included educational programmes and techniques such as the Expert Patient Programme, the Living Well Programme and Cornerstone Conversations. Technological solutions, such as hand-held health records, could engender patient activation by helping them to monitor and self-manage their health. Non-technical solutions included a written record of lab tests and a written dedicated action plan. Small changes might also help empower people. Examples included asking the patient to write down the date of their next appointment or providing the facility for them to book their own appointment.

Interventions that address opportunity: social influences

Key informants emphasised the importance of continuity of care, of health-care professionals building a relationship with patients and making them feel welcome and valued. There should be an ongoing dialogue between consultants and their patients. However, developing and maintaining this relationship could take time.

In-house training could make staff more welcoming and engender a clinic culture of valuing the patient being there. When patients came back to the clinic, it was particularly important that staff should be trained to respond to them in an appropriate and sensitive way.

The importance of peer support in building confidence and understanding was highlighted and different ways of providing peer support were discussed. Peer support could involve patient representatives who were part of the service or peer mentors who were either employed by or affiliated to community organisations. There were different ideas on whether peer mentors should be volunteers or paid members of staff. Being a paid member of staff could raise the issue of confidentiality and discussing sensitive issues about the clinic and the staff with another member of the staff. On the other hand, paid staff could work with other professionals and provide an effective way of facilitating valuable discussion. Payment also implied a professional identity and certain expectations of performance. Professor of Medicine (MM) talked about the high turnover in peer support and how stressful the work could be. He raised the importance of training, roles and responsibilities, and supervision.

When peer support was integrated into the HIV service, it could be used to change the dynamics and start conversations. Peer mentors had time to talk to patients and gave patients a sense of hope:

It’s very powerful, I can say to a patient, shake their hand, say, I’m Chris, I’ve been living with HIV for over thirty years, so any idea of you dying – forget. And they can see that . . . it’s a visual form of education.

Patient representative 3 (CS)

In addition to one-to-one discussions, peer support groups held in the clinic or elsewhere could also enable people to learn from each other. It was suggested that peer mentors could also visit community clinics once a month to provide information.

It was crucial to try and engage with people as soon as possible after diagnosis both through one-to-one peer support and courses for the newly diagnosed. Newly diagnosed people should be routinely directed towards this support and, when peer mentors were available in the clinic, they could be introduced to them right away.

Interventions that address opportunity: barriers

Key informants indicated that multidisciplinary, holistic support helping patients to prioritise their health care should be funded and consistently available across the country. HIV clinics should work towards providing an overarching supportive environment and not just a clinical service, with social support incorporated through the clinic or in collaboration with external agencies, such as community organisations and Citizens Advice Bureaux.

Making changes to patient pathways could mean that patients would not always need to come into the HIV clinic for bloods or to see the doctor; services could be taken to them. The community nurse specialist role needed definition but might deliver such a service, providing continuity of care by seeing patients in the clinic, on the ward and at home. Such changes to service delivery might also include greater involvement of GP practices. Key informants suggested different configurations, including HIV clinicians or nurse specialists doing clinics at the GP practice, bloods being done at the GP practice or at home and skilling up GPs for shared care. Some patients, however, found it easier to have all their care in one service:

You go to your GP and say to your GP, my HIV clinician said I must come to see you about this, and the GP says, I’m sorry, I can’t treat you for that, you need to go back to your clinic . . . we’ve got to get rid of this ping-pong.

Patient representative 1

In addition, key informants suggested that some consultations for some patients could be conducted using various technological and innovative solutions including a virtual HIV clinic; consultations or results by telephone, e-mail, WhatsApp (WhatsApp Inc., Mountain View, CA, USA) or Skype; and electronic access to results.

Patient-friendly facilities and services that were flexible and accessible might facilitate attendance. Key informants talked about organising facilities so that the waiting area was out of earshot of reception, seats were not facing the door, and there was a place for prams and child care provision. A comfortable environment could make a difference and calm the patients. Other suggestions included evening and weekend clinics; a more flexible walk-in system; fewer appointments if patients were stable on treatment; co-ordinated appointments so that patients could see all health-care professionals on the same day; and one telephone number to call. Volunteers from community organisations could also help with logistical issues including babysitting or driving someone to clinic.

Interventions for addressing financial issues might enable patients to stay engaged. Suggestions included giving supermarket vouchers to patients who were badly in need. The problem of getting to the clinic could be addressed by pre-paid Oyster cards, an emergency travel fund or guaranteed travel reimbursement. It was important, however, to explain what was available to patients and facilitate travel reimbursement because it would be embarrassing and complicated to claim. People who were very difficult to engage in care might be provided with a cheap mobile phone to call or text clinic and which the clinic could also use to contact them.

Intervention 1: structured peer involvement

Intervention 2: ‘one-stop-shop’ multidisciplinary team clinic

Intervention 3: running a clinic in a non-NHS/alternative NHS setting

Intervention 4: low-cost clinic-wide intervention

Health costs of disengagement

We examined the association between EIC and mortality, adjusting for CD4 count to see if differences were being mediated by differences in CD4 count profiles. There were 2279 deaths in the cohort and the analysis showed that each 10% increment in EIC was associated with a 9% reduction in the risk of mortality. This remained fairly constant when we adjusted for fixed covariates but, importantly, when we adjusted for the CD4 count at the time and over follow-up, the association was attenuated completely to 1. The results were consistent when we included AIDS or death as the outcome measure, suggesting that patients with poor EIC tended to experience poorer outcomes because their CD4 counts were generally lower.

Next we restricted the analysis to patients who had started ART. The analysis showed that around 40% were engaged in care for 90–100% of the time. As the proportion of EIC increased, more MSM and people of white ethnicity were represented among these groups. As we would expect, patients with higher engagement had a higher CD4 count on average and, as a reflection, were more likely to start a non-nucleoside as opposed to a PI.

The association with mortality was equally strong for this group, even though they were on treatment, with a very strong association between EIC before starting treatment and mortality after starting treatment. This was affected only minimally by adjustment for fixed covariates, but when we controlled for the CD4 count and viral load, the effect moved towards 1 and became non-significant. This confirmed that people with poor EIC generally not only have lower CD4 counts at the time of starting treatment but have poorer CD4 count and viral load responses after starting treatment.

Our analysis indicated that higher EIC is associated with improved clinical outcomes, even when looking at least 1 year into the future and when we consider a subgroup of patients who are already in care and who start treatment, a group that we would think would be at relatively low risk of mortality. On the whole, clinical outcomes were largely explained by poorer CD4 count profiles in those with suboptimal EIC.

Although this analysis is subject to the limitations of the EIC algorithm, as described earlier, it benefits from the advantages of using the extensive UK CHIC data set to explore the health costs of disengaging from HIV care. Although death and AIDS diagnoses occur infrequently in the population, we have found a strong association with disengaging from care. The analysis is also complicated by the fact that these negative outcomes are associated with high levels of engagement in the period leading up to the event when patients attend services because they are unwell. We have accounted for this by looking at EIC before starting ART and death after starting ART in order to avoid contamination of this ‘reverse causality’ effect.

Association between test costs and disengagement from care

Our analysis of financial costs was limited by available data in UK CHIC on NHS resource use, and hence we focused on laboratory test costs. Given the limitations of the data, our analysis was descriptive. There was some suggestion that people in the most engaged groups had more tests per patient in each quarter, and therefore higher test costs and that those who are least engaged seem to have fewer tests and lower test costs but the differences are small. There does not seem to be any appreciable crossover in test costs between pre-ART EIC groups in later quarters as their health declines.

Further research focusing on a wider range of NHS resource use would be beneficial, including clinical appointment costs, medication costs and costs of hospital stays. This requires a data set with detailed information on NHS resource use associated with each of these items plus measures of EIC. A more sophisticated analysis would take a wider costing perspective, such as a societal perspective, which would also consider costs of disengaging from care on patients and families, business and costs to social services and other parts of the public sector. When considering impacts on patients and families, this should account for the possibility that, on the one hand, engaging in care might increase time and travel costs to patients and families (e.g. taking time off work to attend clinics, travelling to and from clinics), but, on the other hand, may also reduce costs to patients and families by avoiding deteriorations in health.

As well as evaluating the costs of disengaging with care, further research would also be beneficial to evaluate the cost-effectiveness of interventions to improve EIC, such as the interventions described in Chapter 7. Economic evaluation of interventions such as this should account for:

• the cost of the interventions

• the impact of the interventions on EIC

• the impact of engaging in care on health outcomes (both in terms of mortality and health-related quality of life)

• the impact of engaging in care on costs associated with illness, from both NHS and wider societal perspectives.

As noted above, evaluation of the interventions on mortality, health-related quality of life and NHS and other costs should appropriately account for biases caused by ‘reverse causality’.