Aim

The aim of the health economic analysis was to compare four models of primary care provision for people who are homeless with respect to: (1) the resources assigned by services to the care of registered patients who are homeless and (2) the use and costs of health and social care services over a 12-month period by the case study participants. It also sought to evaluate outcomes for participants in relation to service use costs. The four models of primary care provision, as described previously, were (1) Dedicated Centres providing care only to people who are homeless and other marginalised groups, (2) Mobile Teams, (3) Specialist GPs providing targeted services (e.g. specific clinics) within their practice and (4) no special provision by practices (Usual Care GPs).

Data sources

People who were homeless were recruited through the services/practices (CSSs) that were involved in the study. Two sites and 96 participants were included for each of the three non-usual care models; 75 people were recruited from four Usual Care GPs; overall recruitment was 363.

Information on the resourcing of the CSSs was requested through interviews with managers and other personnel. Questions were asked about organisation and financing, the patient population and proportion who were homeless, staffing, facilities, dedicated clinics, other services provided (e.g. for mental ill health, drug misuse and lifestyle), and integration with health and social care systems.

Service use data were collected for each participant starting 4 months before the date of baseline interviews to the date of the 8-month interviews, giving 12 months of data by two methods. Self-report data were collected retrospectively by recall through questions embedded in the interview schedules (at baseline and at 4 and 8 months) covering contacts with the primary health care centre (GPs and nurses) and other health and allied health professionals, out-of-hours services (A&E, NHS walk-in/urgent care clinic, NHS 111, ambulance call-out) and hospital admissions. Primary care medical records of participants were also accessed (subject to separate consent) at the end of the 12-month follow-up period. These were taken as the primary source of health service use data as they were more comprehensive and complete than information obtained from participants, and they were also considered to be more reliable. GP records provided service use data of participants in the three GP-led models; both Mobile Team and GP records (when shared) were used for participants in the Mobile Team model. Use of accommodation (e.g. nights in hostels) was not included.

Self-report data were used to provide information on items that were not routinely available from medical records, such as contacts with key workers, substance misuse services, pharmacists for methadone, local authority housing and social workers, personal care workers (e.g. home care workers) and volunteer supporters. Dates were noted when participants were not using local services during the 12-month observation period, for example because they were in prison, were in rehabilitation, had left the area, or had died. Data were recorded initially by 4-month periods, but were amalgamated for analysis. Service use data were costed in 2020 Great British pounds using national validated tariffs;¹⁰⁷ costs of tests were accessed from a variety of sources (see Appendix 7, Table 58).

Service level

Information reported by managers and other personnel regarding patient populations, resourcing and organisation of services is reported in Chapter 5. All sites were commissioned through the NHS and received local enhanced payments for providing services to people who were homeless (UC1 received payments for providing services to patients who had substance misuse problems). The Dedicated Centre, Specialist GP and Usual Care GP sites were GP led and provided care primarily at general practice premises; the Dedicated Centres and Specialist GPs also offered some outreach. Mobile Teams were run by specialist nurses through clinics in day centres and hostels. They provided more limited services around wound dressing, monitoring of long-term conditions and health promotion, referring patients on to other services as needed, including to local GPs with whom they worked closely, and shared medical records. A GP out-of-hours service was available in all sites.

The patients receiving care from Mobile Teams also differed significantly in some respects from those registered with GPs in the other models (see Chapter 6). Mobile Team patients were more likely to have been born outside the UK, to be European Union migrants and to sleep rough; they also were less likely to use drugs.

Participant service use

Missing information

Around three-quarters of participants were available to use local services (rather than being in prison, rehabilitation or abroad) for at least 10 out of the 12 months of follow-up in the three GP-led models; 68% of participants were available in the Mobile Team group. Overall, the distribution of number of days when participants were in local circulation was similar across models (Table 39). Hence, the comparison of service use across the models was based on totals and averages without any adjustment for missing information. Full service use data were unavailable for 25 participants: GP records could not be accessed for 14 participants in the Usual Care GP model (UC1 = 11, UC2 = 1, UC4 = 2), and key worker information was missing at all three time periods for a further 11 participants (DC1 = 1, DC2 = 2, MT2 = 7, UC2 = 1). These individuals were excluded from some parts of the analysis (including Grand Total Costs because self-report data on tests were not collected). A meticulous process of data extraction from medical records (when available) and participant interviews, cross-checking between the two, and quality checks by a second coder meant that missing data was minimal. In many cases, self-report data were available in the absence of medical records.

TABLE 39

Number of participants, by number of days with access to services by model of primary care provision

Contacts with services

A full breakdown of contacts by individual item of service use is given by model and site in Appendix 8, Tables 59–62. The rate of use of many services was low and items were grouped for further analysis (GP, nurse, other community, allied health professionals, substance misuse including smoking services, social, out-of-hours, hospital in/out/day, tests, personal caring, optician, dentist) (see Appendix 9, Table 63).

The number of contacts over a 12-month period with different groups of services, by model of primary care provision, is shown in Appendix 10, Table 64. Other than contacts with smoking, drug and alcohol services, which included daily collection of methadone from pharmacists for some 40% of participants, the most frequently accessed service was the GP (overall median number of contacts, 7: Dedicated Centres n = 14, Mobile Teams n = 6, Specialist GPs n = 7.5 and Usual Care GPs n = 4). Nearly all participants (97.0%) in the three GP-led models had at least one contact with their GP (at the practice, on the telephone or elsewhere), but this was lower (91.7%) in the nurse-led Mobile Team model. Across all models, one-third (33.1%) had hospital admissions and around two-thirds (65.0%) used out-of-hours services. Forty-eight participants had 10 or more contacts with out-of-hours services over the study period (maximum number of contacts was 92). The number of out-of-hours contacts was positively correlated with the number of GP and nurse contacts (n = 363, Spearman’s ρ = 0.302; p < 0.0005). This association was statistically significant in both Dedicated Centres and in one site in each of the other service models (MT2, SP2 and UC3), suggesting that out-of-hours services are not necessarily used as a substitute for GP or nurse consultations and that some participants were frequent users of both (Table 40).

TABLE 40

Associations between GP and nurse contacts and out-of-hours (A&E, NHS walk-in/urgent care clinic, NHS 111 and ambulance call-out) contacts, by CSS

Costs

The cost of contacts with services and professionals is shown in Table 41. Considering all health and social care use (except dentist and optician, and the extensive personal care support received by two participants), the highest-cost items were hospital stays (other than for detoxification) (mean £2164, median £0), out-of-hours care (i.e. A&E, walk-in, urgent care, NHS 111 and ambulance) (mean £792, median £296) and contacts with GPs (mean £686, median £462). The Grand Total Costs of service use were highest in the Dedicated Centre model.

TABLE 41

Costs (£, 2020) of service use over the 12-month study period by model of primary care provision

Average costs were compared between models for six main service use groups using the Kruskal–Wallis test, and between pairs of models using the Mann–Whitney U Test (Table 42). Statistically significant differences existed between models for the following: GP, nurse and HCA contacts (higher for Dedicated Centres than for the other three models; lower for Usual Care GPs than for Mobile Teams and Specialist GPs); use of smoking, drug and alcohol services, including methadone collection from pharmacists (higher for Specialist GPs than for Mobile Teams and Usual Care GPs; higher for Dedicated Centres than for Mobile Teams);

TABLE 42

Differences in Health Service Models between main cost items (£, 2020)

social care contacts with care co-ordinators/managers, key workers, and housing and social care officers (higher for Mobile Teams than for the other three models); Grand Total Costs (higher for Dedicated Centres than for Mobile Teams and Usual Care GPs; higher for Specialist GPs than for Usual Care GPs). No significant differences were found between models for out-of-hours services or for hospital nights (including psychiatric, excluding detoxification). Distributions were right skewed for all items and models. Differences are illustrated in box plots (Figure 5).

FIGURE 5

Comparison of costs by Health Service Model. (a) GP, nurse and HCA contacts; (b) smoking, drug and alcohol services, pharmacist delivery of methadone; (c) social care contacts (care managers/co-ordinators, key workers, housing and welfare officers); (d) (more...)

Comparisons of sites within models were also conducted for each of the six main cost groupings. No statistically significant differences were found except between the two Dedicated Centre sites for costs of participant use of smoking, drug and alcohol services, which were significantly higher in DC1 than in DC2 (data not shown) due to the large numbers of contacts with pharmacists for collection of methadone.

Predictors of service use and costs

The role of service model in predicting use (or not) of out-of-hours services and Grand Total Costs was explored using regression analysis (with forced entry of the service model variable). Understanding the predictors of use of out-of-hours services is of interest because people who are homeless often access care through this route, at higher cost, rather than using general practice. The analysis of Grand Total Costs sought to uncover the drivers of service use among the homeless population. Co-variates are shown in Appendix 3, Tables 48 and 49.

Stepwise logistic regression of use of out-of-hours services found no statistically significant association with Health Service Model. The only significant predictor was number of changes of accommodation during the study period, with each additional change rendering a participant 1.45 times more like to use an out-of-hours service. Hence, for example, compared with a participant with one change of accommodation, a participant with six changes of accommodation would be 1.45 × 1.45 × 1.45 × 1.45 × 1.45 = 6.41 times more likely to use one or more out-of-hours services (Table 43).

TABLE 43

Final model for out-of-hours service use following stepwise logistic regression, with forced entry of Health Service Model

Grand Total Costs were positively skewed and thus transformed to enable a linear regression of log Grand Total Costs. The final model indicates that participants registered in Dedicated Centres incur significantly higher costs than those in Usual Care GPs (Table 44). Higher Grand Total Costs are also associated with spending a higher proportion of the study time in accommodation with staff on site and having more changes of accommodation during the study period; lower Grand Total Costs are associated with being Black or Black British and recently reporting involvement in education/training/employment. Per annum costs vary between £469 and £21,590 for the lowest- and highest-cost scenarios.

TABLE 44

Final model for Grand Total Costs, with forced entry of Health Service Model

Regression analysis of hospital admissions was hindered by the variability in number of admissions by a small number of participants. Reduction of data to a dichotomous variable (0 vs. to 1 or more admission) found no association with models of care and failed to identify any significant predictors.

Costs in relation to outcomes

The rates of use and the costs across all services were significantly higher among participants in the Dedicated Centres than in the other three models, and significantly lower in the Usual Care GP model than in the Mobile Teams and Specialist GPs. The specialist interest of staff for care of people who are homeless, and high levels of integration with relevant services (mental health, substance misuse) and local homeless hostels and day centres, may contribute to the higher service use of participants registered with Dedicated Centres. In contrast, Usual Care GP sites are not primarily set up to deal with the special needs of people who are homeless; they reported lower levels of integration with other providers, and staff may have been less proactive in identifying issues and referring on. Drawing on findings from previous chapters, there is some evidence that higher levels of service use are reflected in better outcomes, but this does not hold for all measures.

Screening activity (the primary outcome) was not significantly different between the three GP-led models, but was significantly lower in the Mobile Team model. Similarly, performance on care for SHCs (secondary outcome, overall score) did not differ between models, except that the Usual Care GP model recorded a lower score than the other three models for people who used substances. Dedicated Centres did, however, score significantly better on the continuity of care component for all four of the SHCs analysed, with Mobile Teams performing least well on this outcome. Controlling for other factors, participants in the Dedicated Centre and Specialist GP models showed small improvements in PCS of the SF-8 over an 8-month period; the well-being of participants (measured using the SWEMWBS) in the Specialist GP model also improved. However, this finding should be viewed with caution: approximately 100 participants were not available for follow-up, and the 250 participants for whom follow-up data were available may not have been representative of the entire group. Although relatively low proportions of participants in all four models reported receiving nutrition advice (mean 28.5%), this was lowest among those in the Usual Care GP model (13.3%). In terms of satisfaction with the primary care service, participants in all three specialist services (Dedicated Centres, Mobile Teams and Specialist GPs) rated their experiences highly (more so than is recorded for the general population); participants in Usual Care GPs rated their experiences worse than the general population did, but this was largely a result of low scores in just one of the sites. Costs and outcomes across the four Health Service Models are summarised in Appendix 11, Table 65.

Summary

The service used most frequently by participants was the GPs (nurses in Mobile Teams) with which participants were registered. The number of contacts with GPs over the 12-month period was considerably higher among the study sample than has been recently observed among the general population [mean 10.7 (SD 11.4) vs. mean 3.74 (SD 1.24), respectively], and similarly for nurses [mean 8.3 (SD 15.2) vs. mean 1.3 (SD 0.78), respectively].¹⁴⁹ Management of dependency problems (smoking, alcohol and drug misuse) was also associated with relatively high average costs, largely attributable to pharmacist time distributing OST, including methadone and buprenorphine, on a daily basis (n = 142, 39.1% of participants). out-of-hours services were accessed by 65% of participants over the 12-month period, with many having multiple contacts. Modelling suggests that out-of-hours service use is associated with having many accommodation changes; other participant characteristics (demographic or health) and model of care were not predictive.

On average, Dedicated Centre participants incurred higher overall service use costs than participants in other models of care. Having more changes of accommodation and spending more time in accommodation with on-site staff also added to cost; people who were black/Black British or in education, training or employment were associated with lower service use costs. The higher service use in the Dedicated Centre model was associated with significantly better outcomes for some, but not all, indicators measured in this study. Participants reported low levels of satisfaction with Usual Care GP provision and high levels of satisfaction with the other three models.

Limitations

The analysis is limited in various ways. It was not possible to identify resources devoted to care of patients who were homeless within services for calculating an average cost of provision in different models, but all services received NHS payments in recognition of their care of this group. Some service use data were not available from medical records for some participants for all or part of the follow-up period, resulting in differential reliance on self-reported information, which may have been less reliable. We did not adjust for missing information or undertake imputation because the distribution of number of days that participants had access to services was similar across models, which means service use may underestimate actual annual service use across the board. The uneven distribution of missing medical records across sites (higher for Mobile Teams and Usual Care GPs) may have affected the comparison of Grand Total Costs across models. Differences between the mode of delivery of care and characteristics of the patient populations of the nurse-led Mobile Team model, compared with the three GP-led models, need to be borne in mind in the interpretation of comparisons. Similarly, some sites provided more in-house services (e.g. support for smoking and drug and alcohol misuse), which were included in GP and nurse contacts, rather than appearing as a contact with a specialist service. Although this would not affect the analysis of Grand Total Costs, it may have affected comparisons of costs between models. Granularity was lost in the interests of parsimony in the combining of variables for the cost analysis.