An adapted algorithm for patient engagement in care for young people living with perinatal HIV in England

High engagement in care (EIC) is recognised as a crucial step in improving the outcomes of people living with human immunodeficiency virus (HIV) [1‐4]. Adult studies have found a higher likelihood of HIV viral suppression and improved CD4 cell counts in people who have better EIC [5, 6] and higher EIC has been associated with lower healthcare costs [4, 7, 8]. Furthermore, transmission of HIV may be more frequent in people who are not engaged in care [9, 10].

A number of studies have found that young people (YP) living with HIV have worse EIC compared to adults and children [5, 9, 10]. In a national study of 72,218 adults living with HIV in the UK, those aged 15–24 years were the most likely to be lost to follow-up [11]. In another study of 87,146 people living with HIV across all age groups in New York, a U-shaped relationship between age and EIC was found, with people at the youngest (birth to 12 years) and oldest (60 years and older) age ranges having the best EIC, and YP aged 20–29 years the lowest, across the whole age spectrum [9].

However, there is considerable variability in how these measures are defined, in terms of the types of visits considered as either missed or attended (e.g. appointments with the doctor, nurse, psychologist or phlebotomist), and visit frequency (e.g. one visit in 6 months or one visit in a year). There is also a wide variation in the health status of people living with HIV, and those with worse clinical indicators are likely to be scheduled to be seen more frequently. For these patients, use of a simple EIC measure such as a visit frequency of at least one visit in a 6 month period might be obtained but be misleading, and fail to capture missed visits within the period.

Additionally, measuring EIC across the transition period is further complicated because of differences in data sources and guidelines. YP living with perinatal HIV (YPLPHIV) may receive HIV care in paediatric as well as adult HIV services. This requires linkage of paediatric and adult datasets, which may be held separately and by different data controllers. Additionally, whilst many YP seen in paediatric clinics have clinic visits every three to four months [12], older patients who are stable on antiretroviral therapy (ART) may have clinic visits every 6 months, in accordance with adult HIV guidelines [13]. Thus application of a simple definition of EIC to YPLPHIV, with the same length of gap applied between visits, would not account for these nuances and changes over time including transition to adult care.

In this paper, our aim was to account for the complexity of EIC in YPLPHIV by developing a new measure. We describe how we took a published EIC algorithm for adults with HIV, originally developed by Howarth et al. [14], and adapted it for use in YPLPHIV in England. Howarth et al.’s rationale for developing this new algorithm was that standard EIC measures do not take into account how the frequency of clinic attendance is based on individual patients’ health and treatment status and that this changes over time [14]. Their algorithm used clinical factors to predict when adult patients should next be seen in clinic (i.e. the predicted next appointment date). We updated Howarth et al.’s definitions to more recent adult treatment guidelines and modified appointment scheduling for young people still in paediatric care based on the latest European paediatric guidelines [12, 13]. This adapted algorithm was then applied to a dataset of YPLPHIV in England in the Adolescents and Adults Living with Perinatal HIV (AALPHI) cohort [14]. We hope that findings from this work can be used to help create a risk assessment that can be used to identify those at higher risk of disengaging from care for additional support.

The AALPHI cohort study was a prospective study in which a wide range of data were collected to describe the impact of life-long HIV and long-term ART on health outcomes. YPLPHIV (aged 13–21 years) were recruited from NHS clinics and voluntary sector organisations across England, and were interviewed at two different time points between 2013 and 2017. Detailed AALPHI methods have been described previously [15, 16]. Ethical approval was obtained from Leicester Research Ethics Committee. All AALPHI participants living with PHIV were also in the national UK Collaborative HIV Paediatric Study (CHIPS) [17], which captures the clinical data used in this analysis.

The Howarth algorithm was updated to the most recent British HIV Association (BHIVA) clinical monitoring guidelines at the time, [13] and then reviewed alongside the Penta paediatric HIV treatment guidelines [12]. Where Penta guidelines recommended a shorter maximum time to next appointment for a group of patients, the algorithm was updated accordingly [17]. The adapted algorithm was then reviewed with two clinicians caring for YPLPHIV in large London clinics, as well as the Steering Committee of CHIPS. Three main groups of YPLPHIV were identified. Within each group, a decision tree approach was taken to predict the next scheduled appointment date, with the following variables used to predict the next appointment: viral load (VL), CD4 cell count, CDC C (AIDS defining) events, weight, ART start/new regimen, number of ART drugs, time on ART, third agent (protease inhibitor (PI)/non-nucleoside reverse transcriptase inhibitors (NNRTI)), care provision (paediatrics/adult).

Of 316 patients in AALPHI, 41% (n = 129) were male, 82% (n = 258) of black African ethnicity and 58% (n = 184) were born outside the UK/ Ireland (Table 1). Twenty seven per cent of participants (n = 84) had a previous CDC C event. At baseline (time of AALPHI interview), median [IQR] age was 17 [15-18] years, median CD4 count was 597 [427, 791] and 69% (n = 202) had viral load ≤ 50c/ml.

Across the 12 months following the AALPHI interview, 10 participants were excluded due to having no clinical (CHIPS) data. For the remaining 306 participants, there were 3,585 months of follow-up, in which a clinic visit was recorded in 34% (n = 1,204) months. Following imputation, 3% (38/1,204) visit months remained unclassified due to incomplete ART and/or clinical information. These 38 were coded as in care for one month (the minimum time to next scheduled appointment). If the participant did not attend within this month, then subsequent months without a visit were dropped until the participant attended again. The remaining 1,166 visit months were classified into three main groups as follows:

The flowcharts for groups A-C show the clinical decisions used to predict a participant’s next appointment with timings of the next scheduled appointment indicated in all the shaded terminal nodes.

In this paper, we describe how we adapted an existing EIC algorithm for adults living with HIV for YPLPHIV. Penta paediatric and BHIVA adult HIV treatment guidelines were used alongside expert opinion to modify and develop the flowcharts for this population. These adaptations increased the number of different pathways to determine the next predicted appointment from 15 in the adult algorithm to 37 for young people. Although ultimately more complex than the adult algorithm, the YP algorithm is a novel framework which we then applied to a large observational dataset, providing valuable data on EIC.

Engagement in care is commonly measured as a threshold of a specific number of visits per year, and it is generally assumed that a higher number of visits equates to better engagement in care [20‐22]. Some studies have simplified this to, for example, one or more visits in a given year, [21, 23] which gives a population estimate useful for public health monitoring and the cascade of care. However, these approaches do not consider the varied and changing clinical scenarios of patients. For a young person who is virally suppressed on ART, two clinic visits per year may indicate an entirely appropriate level of engagement in care. For a young person who does not take their ART and has a low CD4 count, or a young person not yet weighing 40 kg (and therefore still requiring ART adjustment based on weight), attending two visits in a year would indicate sub-optimal engagement because they should be seen in clinic more regularly. Our model is novel because for the first time it incorporates changes in HIV treatment guidance between paediatric and adult care and considers the clinical scenario of children and young people in the EIC measure. However, it does highlight the importance of linking YPLPHIV transitioning from paediatric to adult care in surveillance data sets (e.g. in the England setting from the Children’s HIV and AIDS Reporting System (CHARS) to the HIV and AIDS Reporting System (HARS)) to enable this work to be implemented moving forward.

Findings from this study suggest that most patients followed the scheduling requested by the clinician and were engaged in care. However, findings also demonstrated that participants did not stay on the same appointment schedule, with 37% of young people having at least one of their appointments predicted at 1 month, 40% at 2 or 3 months, 61% at 4 months and 24% at 6 months. This is an important public health message because it shows that it is hard to predict the care needs of these young people, as their needs may change over time. Our analysis only considered one year of follow-up, and if the analysis was over a longer period of time, or if indeed the whole of adolescence was considered, the pattern of appointment scheduling would likely vary even more. Furthermore, this finding supports the importance of a flexible measure of EIC that can pick up the nuances of clinical care in this age group.

In our study, we also observed that young people were most commonly predicted an appointment in 4 months’ time but that the median number of appointments was 4 visits per patient per annum (i.e. 3 monthly). This is most likely explained by the early attendance shown in the results. Reassuringly, when characteristics of participants who attended early were compared to those who did not, no differences were found. This early attendance therefore raises a question about the accuracy of the algorithm and whether doctors suggested visit scheduling less frequently than was actually undertaken in clinical practice. Alternatively it could suggest that doctors think that patients are able to be seen at longer intervals but that patients themselves wish to come earlier. Comparisons across appointment schedules should be viewed with caution because there is greater opportunity for participants to be early for a 6 month appointment than an appointment predicted in 3 or 4 months’ time. However, the reasons for early attendance could be further investigated as early attendance may have implications for costings and resources if people are being seen more often than needed based on their clinical status.

It is hard to directly compare our EIC findings to other studies due to considerable variability in how EIC is defined, in terms of the types of visits considered as either missed or attended (e.g. appointments with the doctor, nurse, psychologist or phlebotomist), and visit frequency (e.g. one visit in 6 months or one visit in a year [24]). In addition, many decisions underpinning EIC definitions are ultimately pragmatic and based on the clinical and appointment information available to researchers, as well as simplicity of approach to analysis. Nevertheless, it is important to try to contextualise findings. In our study, 87% of patient months of follow-up were defined as in care. European studies of YPLPHIV have reported engagement ranging from 80 to 98%, [23, 25, 26] and sub-Saharan African studies 83% to 98%, with studies commonly using a criterion of ≥ 1 visit in 6 or 12 months across varied age ranges [27‐31]. Only one study has previously estimated EIC in YPLPHIV in the UK. Chappell et al.found that 98% of paediatric patients in the CHIPS study were engaged in care, defined as having a clinical visit, CD4 or viral load measurement, or ART change, within 2016 [23]. The median age was 14.4 years [IQR 11.2, 16.4], younger than in our study, and all children had to not be lost to follow-up in the previous three years to be included in the analysis.

The proportion of participants engaged in care in studies from the USA vary more than studies from Europe and sub-Saharan Africa, largely explained by the difference in EIC measures. Hussen et al. [22] found 56% of young people were engaged in care using a stricter multiple visit marker measure (≥ 2 visits with ≥ 3 months apart within 1 year), compared to Gray et al. [21] who found 99% of young people were engaged in care using a single visit measure (≥ 1 visit within 1 year).

The importance of a flexible EIC measure also has an increasing global relevance due to the roll out of differentiated service delivery across much of sub-Saharan Africa [32]. Differentiated service delivery is a person-centred HIV care approach that aims to increase patient choice around their interaction with the healthcare system, for example by providing options to attend community-based HIV care and/or reduce clinic visits by accessing multi-month dispensing of ART (compared to the more common model of monthly clinic visits for all) [32], However there remain limited data on the impact of differentiated service delivery on EIC and appropriate measurment will need to move away from simplistic measures to provide critical information for clinical care and policy globally. Currently very few countries have linked individual patient data across paediatric and adult health services that would allow this analysis to be carried out. Indeed in England, these data are not currently available outside of the AALPHI study, which is now out of date. We are currently working towards linking paediatric and adult surveillance datasets to allow future analyses across the lifespan. However, the process is lengthy due to complex governance issues.

A key question moving forward in the post-COVID landscape for all EIC measurements and studies is whether we are measuring what we need to measure. Many clinics now have a proportion of visits which are virtual, and this has implications for studies that do not collect data on dates of virtual visits, or indeed actual face to face visits, and instead rely on proxy markers such as viral load and CD4 counts to infer attendances (as in CHIPS). Clinicians, researchers, service providers and commissioners need to work together to ensure the collection and availability of the right data are possible to be able to measure this key HIV outcome.

Despite these limitations, these flowcharts have allowed detailed analysis of routinely collected healthcare data. If this model was regularly applied to surveillance data, results could be used to assess workload in clinic, predict costing, review resources allocation, benchmark between services and form a basis for future interventional studies. The adaption of the flowcharts provides a new approach to measuring EIC in YPLPHIV that is easy to understand and is reflective of changing patient needs over time. However, we acknowledge the complexity of these flowcharts would require ongoing investment which is a challenge in many settings.