Introduction

Programa Maior Cuidado (PMC) is a public sector intervention in Belo Horizonte, Brazil that provides lay carers to vulnerable older people for up to 20 hours per week. Lay carers are trained to meet the basic health and care needs of participants, as well to improve communication with the health and social care systems. Additionally, PMC offers respite for family carers from intensive care responsibilities^1–3.

Prior research showed positive associations between being included in PMC and a greater ratio of planned outpatient visits relative to unplanned ones. This is a desirable outcome, since unplanned visits are less easy for health care providers to manage. Also, they are more likely to occur due to acute health emergencies, whereas planned consultations are more usually for preventive services, such as vaccination. Similarly, research showed positive associations between being included in PMC and a greater ratio of visits made for rehabilitation, as opposed to other reasons⁴.

In this study, we focus on associations between PMC and four outcomes of interest for hospitalised older people: probability of death, probability of intensive care unit (ICU) use, length of stay and total cost of hospitalisation. We interrogate two theories of change:

• PMC promotes timely intervention by anticipating health problems, thus reducing admissions for emergency and acute care, which in turn are associated with higher probability of death and ICU use. Lower rates of ICU use are in turn associated with lower cost.

• PMC facilitates hospital discharge by increasing confidence in community support for older patients, reducing length of stay and total cost.

 
Data and Methods

We had access to the PMC database since the programme began operation (April 2011) through to December 2020. We were able to gather information from the Ministry of Health of Brazil for hospitalizations over the same period of time for Belo Horizonte city⁵. Through a common identifier, we found 588 distinct members of PMC had been hospitalised in Belo Horizonte over that period. Using propensity matching and inverse weight matching^6,7, we created comparison units similar to treatment ones (in this case, being a member of PMC), by employing probabilities (Propensity scores) and weights (Coarsened exact matching). In the case of propensity scores, we use nearest neighbour specifications with ratios of 1:1 and 1 treatment for 5 controls, and a calliper of .1 standard deviations. We used the following contextual variables to create the comparison group: gender, age, race, treatment complexity, and severity classification of reason for hospitalization. Those variables allow us to identify sociodemographic and clinical similarities between the PMC and non-PMC cases.

Finally, using multivariate regression models, we calculate the effect of being in PMC in 4 variables of interest: the probability of death, the probability of ICU use, days of permanence and the total value of hospitalization. This a robust modelling strategy that allows us to control for selection bias at both the treatment and outcome stages, a property commonly referred to as “doubly robust”⁸.

Results

Figure 1 shows the frequency of people enrolled in PMC (the upper chart) and the number of hospitalizations of PMC older people (the lower chart). In both cases, we present information disaggregated by enrolment status (enrolled are shown in green, non-enrolled in red). We observe that from 2019 (the dashed yellow vertical line) the number of hospitalizations of older people not enrolled in PMC spikes, while the number of enrolled older people does not. This suggests that the first wave of the COVID-19 pandemic did not significantly affect PMC hospitalisation numbers.

Figure 1: Time series for hospitalisations

Figure 2 shows the balance of covariates after the matching process for the 3 different specifications (Coarsened Exact Matching, Nearest neighbour, and Nearest neighbour with a ratio of 1 treatment for 5 controls). We observe, in all specifications, a balance in the matching/weighting, whereby the standardized differences in each variable between the PMC group and the control) are smaller than .1.

Figure 2: Covariance balance – different specifications

Our regression models suggest statistically significant associations between enrolment in PMC and the four variables of interest probability of death, probability of ICU use, days of permanence and total value of hospitalization. These are observed across all the specifications. In order to simplify the presentation, we only provide coefficients for CEM models (Table 1). Additional model results can be found in the appendix (see Table 2). In the first two cases, the results are presented as relative risk (compared to not belonging to the PMC). We find across models that belonging to the PMC reduces the risk of death by 3% (95% CI 2%,4%), and reduces the risk of using ICU by 20% (95% CI 17%,23%). In the other two models, results are presented as the difference between being enrolled or not being enrolled to PMC. Across all specifications, enrolment reduces the average number of hospitalization days by 0.22 days (95% CI -.08,-.35), and reduces the average cost of each hospitalization by 375 reais (current value) (95% CI -337,-415), which was equivalent to approximately US$100 per admission (in October 2022).

Table 1: Regression models – CEM – 4 outcomes of interest

INCIDENCE RATIO – DEATHS 

Method: CEM

INCIDENCE RATIO – ICU USE 

Method: CEM

LENGTH OF STAY

Method: CEM                                                             

VALUE OF HOSPITALISATION (REAIS)

Method: CEM                                                             

Discussion.

This study has a number of important limitations. Although matching techniques enable balancing of observations based on observed covariates, unmeasured confounding variables may still be present in our analysis. We address any hidden bias by comparing post-match covariates between groups, and these indicate that the estimates of treatment effects we report are robust. Second, we cannot dismiss possible spillover effects of PMC on the wider health system, which could occur under conditions of limited availability of inpatient hospital capacity relative to need. Finally, we are not able to assess and compare the potential use of private hospitals as there is no public data available for these.

Despite these limitations, the study provides preliminary evidence that, for hospitalised older people, being enrolled in PMC was significantly associated with a reduced probability of death and of ICU use during the hospital stay, as well as a shorter length of stay and a lower average cost of hospitalisation. This represents a contribution to the limited evidence on the effects of broadly similar interventions on outpatient service use^9–11. These findings accord with a separate qualitative study (under review) of PMC which finds some evidence to support the theories of change set out in the introduction. However, the same qualitative study reports that some of these causal pathways are less clearly evident. For example, limited communication between hospital staff and PMC teams is likely to restrict the capacity of PMC to promote timely discharge.

REFERENCES

1. Sartini, C. & Correia, M. Programa maior cuidado: Qualificando e humanizando o cuidado. Pensar/BH Politica Social 31, 10–13 (2012).
2. Lloyd-Sherlock, P. & Giacomin, K. C. Belo Horizonte’s pioneering community care programme for older people | Global Platform. https://corona-older.com/2020/11/24/belo-horizontes-pioneering-community-care-programme-for-older-people/ (2020).
3. Aredes, J. de S., Billings, J., Giacomin, K. C., Lloyd-Sherlock, P. & Firmo, J. O. A. Integrated Care in the Community: The Case of the Programa Maior Cuidado (Older Adult Care Programme) in Belo Horizonte-Minas Gerais, BRA. International Journal of Integrated Care 21, 28 (2021).
4. Lloyd-Sherlock, P., Giacomin, K. & Sempé, L. The effects of an innovative integrated care intervention in Brazil on local health service use by dependent older people. BMC Health Services Research 22, 176 (2022).
5. DATASUS – Ministério da Saúde. https://datasus.saude.gov.br/.
6. Dehejia, R. H. & Wahba, S. Propensity score-matching methods for nonexperimental causal studies. Review of Economics and statistics 84, 151–161 (2002).
7. Iacus, S. M., King, G. & Porro, G. Causal inference without balance checking: Coarsened exact matching. Political Analysis 20, 1–24 (2012).
8. Li, J., Handorf, E., Bekelman, J. & Mitra, N. Propensity score and doubly robust methods for estimating the effect of treatment on censored cost. Stat Med 35, 1985–1999 (2016).
9. Sempé, L., Billings, J. & Lloyd-Sherlock, P. Multidisciplinary interventions for reducing the avoidable displacement from home of frail older people: a systematic review. BMJ Open 9, e030687 (2019).
10. Zurlo, A. & Zuliani, G. Management of care transition and hospital discharge. Aging Clin Exp Res 30, 263–270 (2018).
11. McGilton, K. S. et al. Understanding transitional care programs for older adults who experience delayed discharge: a scoping review. BMC Geriatr 21, 210 (2021).