The effect of patients’ socioeconomic status in rehabilitation centers on the efficiency and performance

Abstract

BACKGROUND

Patients’ socioeconomic status on hospitals’ efficiency in controlling for clinical component characteristics may have a role that has few been studied in rehabilitation centers.

DESIGN

Because of the national health insurance system, rehabilitation centers are free of charge. To answer whether a patient’s socioeconomic status (SES) is associated with efficiency and performance, we use a counterfactual analysis to get the patient’s SES effect “as if” the patient’s case was identical to whatever hospital. We restrained the data to patients from public acute care units where the decision on rehabilitation sector admission is based on availability, limiting bias by confounding factors. Besides, an analysis of six pathologies led to the same results.

SETTING

An exhaustive, detailed administrative database on rehabilitation center stays in France. To define the patients’ socioeconomic status, we use two sources of data: the information collected at the time of the patient’s entry into rehabilitation care and the information collected during the patient’s stay in acute care. This double information avoids possible loss of socio-economic details between the two admissions.

POPULATION

Patients recruited were exhaustively admitted over the year 2018 for stroke, chronic obstructive pulmonary disease, heart failure, or total hip replacement in France in the acute care unit and then in a rehab center. Mainly the elderly population. Information on patients’ demography, comorbidities, and SES are coded due to the reimbursement system. Different dimensions controlling for factors (hospital ownership, patient clinical characteristics, rehabilitation care specificities, medical staff detailed information, and patients’ socioeconomic status), were progressively added to control for any differences in baseline data between the two groups.

METHODS

We assess rehabilitation centers’ efficiency by combining selected outcome quality indicators (Physical score improvement, Cognitive score improvement, Mortality, Return-to-home). The specific Providers’ Activity Index is used to get the performance index.

CONCLUSIONS

The performance of healthcare institutions is correlated not only to the case mix of their patients but also to the socioeconomic status of the patients admitted. The performance needs to be seen in light of patients’ socioeconomic status.

CLINICAL REHABILITATION IMPACTS

The data reveals that patients’ socioeconomic status affects rehabilitation care efficiency and performance. In controlling patients’ socioeconomic status, for-profit rehabilitation hospitals seemed more efficient than public ones.

The search for efficiency introduces incentives to screen patient admissions according to socioeconomic status (SES) correlated to additional discharge constraints and poor underlying health. The question of SES and its impact on a hospital’s efficiency and the clinical components’ interaction is primary when addressing equality in a healthcare system. This paper proposes to assess whether SES affects a hospital’s efficiency and performance.

In the literature, many papers discuss the association between the SES and hospital efficiency and hospital performance.^1-9 Some focus mainly on the length of stay in the acute care unit. There is controversy over whether poor inpatients use more resources and whether hospitals that provide care to the poor deserve extra payment under case-based prospective payment systems. In the paper by Epstein et al.,¹⁰ it is shown that after excluding outliers and adjusting for the diagnosis-related group (DRG), they found that the patients of the lowest SES had hospital stays 3 to 30 per cent longer than those of patients of higher status differences varying with the hospital and the indicator of socioeconomic status. In a Michel et al. paper,¹¹ patient SES appears statistically significantly associated with increased LOS and cost in French hospitals with pediatric departments.

Not done so far to our knowledge, this study focuses on rehabilitation centers that admit patients after acute care stays. The role of these centers is to bolster ability after an acute care stay. They serve to foster a better quality of life after negatively impacted health. According to a study by Bachmann, successful rehabilitation among older patients can improve function, prevent permanent admissions to nursing homes, and decrease mortality.¹²

A choice set of quality indices is used based on the literature. Frequently used in the literature, “returning home after rehabilitation hospitalization discharge” is considered an indicator of successful rehabilitation and thus an indication of the quality of care.^{13, 14} However, some studies show that after a rehabilitation hospitalization, some patients can no longer return to their initial living arrangement and must be readmitted to rehabilitation care facilities, regardless of the quality of care provided by the healthcare structure.^15-17 Therefore, the authors also use the in-patient mortality rate (as other quality indicators), which is not a quality measure per se but rather an outcome only partially determined by healthcare quality. As Urimubenshi¹⁸ discussed, some papers report little or no association of key performance indicators with a lower risk for mortality. Indeed, the mortality varies with the severity of the illness. There is some heterogeneity in severity across patients. A recent exception is Colla,¹⁹ which studies the effect of hospital competition on heart attack, hip and knee replacement, and dementia. However, adherence to common key performance indicators was consistently associated with a lower risk of death after a stroke. Moreover, authors commonly relate this mortality rate index to the quality of services. Schnitzler²⁰ evaluated the functional benefit of specialized neurological rehabilitation versus a general or geriatric rehabilitation (GRC) service. For this paper, we use a set of quality outcome indicators, e.g., physical and mental dependence progress, discharge home, and in-death indicators, to capture the multidimensionality of the healthcare quality. These metrics reveal different quality aspects. As in the recent paper by Johansen,²¹ we use the Mainz²² definition of the three quality domains: structure, process, and outcomes. Therefore, we control for “facilities”, “equipment”, “staffing,” and “multidisciplinary competence”, and we analyze outcome measurements of healthcare quality.

Efficiency differences in rehabilitation units may be due to differences in the clinical patient case mix or SES composition. We approach this issue using an exhaustive, detailed administrative database on rehabilitation center stays in France. The database contains information regarding the patient’s demographic characteristics (age, gender) and the principal diagnosis and comorbidity factors based on the International Classification of Disease, 10th Revision. In this paper, we control for demographic variables and the Charlson index at the patient level. We also control for the patient’s SES, as indicated by deterioration in the level or absence of housing, loneliness (either living alone or declared to be isolated by the medical staff), the inadequacy of social coverage as reported by the medical staff, and the presence of chronic disease. In addition, we control for the French Ecological Deprivation Index (EDI) of the local area (postcode) of home localization. An expert scientific group set up by the French Health Ministry has defined a Providers’ Activity Index (PAI) that is a proxy for the treatment intensity. This paper assesses the performance by combining the outcome quality indicators and the PAI, defined as better quality outcomes through less intense care, i.e., lower PAI.

Pathologies and acts studied are strokes, chronic obstructive pulmonary disease (COPD), heart failure, and total hip replacement. This paper proceeds as follows: we present the database and the methodology used, then analyze the results. Counterfactual simulations are detailed, followed by sensitivity tests. Lastly, there is the conclusion and a discussion of the findings.

Materials and methods

Ethics committee approval statement

The extraction and analysis of the data for scientific research have been conducted with permission from the Hospital Statistics Agency (ATIH), which is the responsible authority for this data. The study was based on routinely collected de-identified administrative data regulated by French law. Data for this study are reported to the National Data Protection Authority. The ethics committee _ CESRESS approved the study under the committee’s reference number: CNIL2019-100001-166-140.

Study population

A population-based retrospective study of patients admitted to a rehabilitation center after acute care unit stays from all French hospitals in 2018 was conducted. This observational study looks at collected archived data at a single point in time with no follow-up prospective data collected. In clinical research, this study is usually used mainly to understand the prevalence of a disease with randomized controlled trials, which match the groups and control for various variables.²³ In economic papers and papers on economic issues comparing the efficiency and performance of interventions between hospitals, what is usually used is a population-based retrospective study.^24-30 We then assess differences using counterfactual analysis on an exhaustive, detailed administrative database that allows us to get more statistical significance, given the study’s statistical power.^31-33

The database comes from the French National Hospital discharge diagnosis databases: PMSI-MCO for acute care and PMSI-SSR for rehabilitation care. Both databases contain information regarding the patient’s demographic characteristics (age, gender), primary diagnosis, comorbid factors, and SES based on the International Classification of Disease (ICD), 10^th Revision. Patients are followed throughout their hospital stays using an anonymous identification number unique to each patient. Patients first identified in the acute care database were patients with a primary diagnosis or occurrence of stroke (respectively, COPD, heart failure, or total hip replacement). They were then searched for the Rehabilitation Database within 0 to 3 days after acute care discharge using the anonymous identification number and the date of discharge from acute care stay. Data selection is patients admitted to a rehabilitation center between January 1 and December 31, 2018, and discharged between 2018 and 2019.

Data sources and variables

The databases were obtained from the national hospital information agency (Agence Technique de l’Information sur l’Hospitalisation – ATIH). The extraction and analysis of the data for scientific research have been conducted with permission from the hospital statistics agency (ATIH), which is the authority responsible for this data. Informed consent is not required since the study was based on routinely collected de-identified administrative data regulated by French law. Data for this study are reported to the National Data Protection Authority (CNIL number 2019-100001-166-140).

To define the patients’ SES and clinical characteristics, we use two sources of data: the information collected at the time of the patient’s entry into rehabilitation care; and the information collected during the patient’s stay in acute care. This double information avoids possible loss of socio-economic details between the two admissions. Besides, using the information on patients from acute care public hospitals avoids coding practice heterogeneity between ownership in rehabilitation centers, which could affect results.

Several variables were computed. The Charlson Index was calculated using the comorbid factors stated in the acute care database and adapted to the study: pathologies relating to this index were taken into account according to published algorithms,^{34, 35} with the corresponding weightings (see Table I for detailed information).

Table I. —Preliminary statistics for different outcomes, used as indicators of performance.

Pathology/act		Stroke	COPD	Heart failure	Total hip replacement
Number of observations
	Whole set	30,585	27,421	34,284	33,205
	For-profit	29.1%	38.7%	33.3%	51.8%
	Non-profit	23.7%	33.3%	16.6%	23.3%
	Public	47.3%	28.0%	50.0%	24.5%
Charlson’s index (score from 0 to 2)
	Whole set	2.91	2.61	3.07	1.85
	For-profit	2.87	2.45	2.95	1.80
	Non-profit	2.88	2.55	2.97	1.84
	Public	2.95	2.86	3.05	1.94
Patient’s SES (score from -6 to 6)
	Whole set	0.006	-0.063	-0.065	-0.083
	For-profit	-0.126	-0.072	-0.296	-0.226
	Non-profit	-0.019	-0.16	-0.181	-0.148
	Public	0.095	0.071	0.127	0.278
Age (years)
	Whole set	70.5 yrs	69.5 yrs	83.4 yrs	74.8 yrs
	For-profit	71.2 yrs	70.3 yrs	82.3 yrs	74.8 yrs
	Non-profit	66.2 yrs	65.1 yrs	82.3 yrs	74.2 yrs
	Public	72.3 yrs	73.6 yrs	84.6 yrs	75.5 yrs
Physical dependence score improvement (from the admission to the discharge)
	Whole set	1.498	0.399	0.543	1.509
	For-profit	1.834	0.581	0.881	1.619
	Non-profit	1.618	0.272	0.510	1.528
	Public	1.241	0.298	0.331	1.262
Cognitive dependence score improvement (from the admission to the discharge)
	Whole set	0.266	0.032	-0.0206	0.154
	For-profit	0.371	.093	0.125	0.189
	Non-profit	0.267	0.004	-0.071	0.162
	Public	0.205	-0.022	-0.099	0.074
In-death
	Whole set	0.023	0.022	0.069	0.001
	For-profit	0.016	0.017	0.047	0.001
	Non-profit	0.014	0.014	0.050	0.001
	Public	0.032	0.038	0.089	0.001
Discharged home
	Whole set	0.628	0.769	0.600	0.892
	For-profit	0.632	0.799	0.640	0.899
	Non-profit	0.653	0.797	0.633	0.891
	Public	0.614	0.697	0.563	0.880
Provider Activity Index (PAI)
	Whole set	103.262	62.222	34.716	68.489
	For-profit	105.165	66.031	39.255	68.757
	Non-profit	112.850	64.064	32.209	65.127
	Public	97.296	54.760	32.528	71.085

The Charlson Index predicts ten-year mortality using 22 comorbidity conditions. Each condition is scored a 1, 2, 3, or 4, depending on the severity of the condition, and is calculated based on all diagnoses recorded in hip replacement admission. We group patients into five categories: a score of zero for no comorbidities; a score of 1 for “very moderate comorbidities”; a score of 2 for “moderate comorbidities”; a score of 3 for “severe comorbidities”; and a score of 4 for “very severe comorbidities.”

From a specific rehabilitation DRG group composed of three levels according to severity, we define the severity level as the highest. A French EDI (European Deprivation Index), developed by Pornet,³⁶ is calculated at the postcode level. The level of deprivation rises with the French EDI values.

We mobilize different outcomes: in-patient death, discharge home, and dependence scores. For dependence scores, we compute two aggregated scores:

• The physical dependence score is from an aggregated list of items divided into four categories concerning activities of daily living (ADL): 1- dressing and personal hygiene, 2- transfers/bed mobility, locomotion, 3- eating, and 4- bowel and bladder control. The value is administratively recalibrated from 0 to 3.

• The cognitive dependence score is from an aggregation of two factors: 1- communication, 2 - behavior. The value is administratively recalibrated from 0 to 3.

This information is available both at the patient’s admittance and discharge. The difference between these values is calculated for each score. We interpret this difference as the dependence score improvement. The higher the value, the more significant the improvement.

The data also contains a providers’ activity index (PAI) regarding rehabilitation and re-education providers’ activity as set up by a group of scientific experts coordinated by the ATIH (an agency for the Health Ministry). It measures the provider’s staff activity for patients during their rehabilitation stay. This index, combined with a series of other indicators, is currently used to increase efficiency and effectiveness in rehabilitation centers’ reimbursement system reform. It is computed routinely according to the type of interventions by physiotherapists and other medical and non-medical staff. Each intervention is weighted by two specific classification guides (the CSARR and CCAM guides). The PAI describes the patient care provided, not the patient’s health status. It is assessed in time and is not limited in value. The higher the PAI, the more intense the healthcare provided to the patient.

Disorders selected are frequent disorders with a rehabilitation stay after an acute care stay. Three of them are emergency disorders. Total hip replacement is an elective procedure. Patients admitted to rehabilitation centers are, on average, over 80 years old.

Methods

As dependent variables, the quality indicators are composed of quantitative and qualitative variables. We used the OLS model as outcomes (dependence scores) for quantitative dependent variables. We conducted two multivariate analyses using logistic regression: one analyzed factors associated with in-hospital death, and the other examined the factors associated with a discharge home.

The models are run, with Y^d being one of the dependent variable outcomes, defined as follows:

1. Case 1: Y^d=f(ownership/ u₁).

2. Case 2: Y^d=f(ownership, age group: less than 4, 5-24, 25-59, 60-69, 70-74, 75-79, 80-84, 85-90, over 90, Charlson Index, severity, age group crossed Charlson Index, age group crossed severity, the presence of chronic disease/ u₂).

3. Case 3: Y^d=f(ownership, age group, Charlson Index, severity, age group crossed Charlson Index, age group crossed severity, the presence of chronic disease, provenance: nursing home, residence as well as entrance from emergency unit, the type of specialization of the rehabilitation center: musculoskeletal disease, nervous system disease, cardiovascular disease, respiratory disease, endocrine and digestive system, oncology/hematology, severely burned patients, addictive behaviors, geriatrics, multi-purpose / u₃).

4. Case 4: Y^d=f(ownership, age group, Charlson Index, severity, age group crossed Charlson Index, age group crossed severity, the presence of chronic disease, provenance: nursing home, residence as well as entrance from emergency unit, the type of specialization of the rehabilitation center, and rehabilitation center information: type of health staff in percentage and in number of activities per stay (family doctor, cardiologist, endocrinologist, neurologist, pediatrician, pulmonologist, specialist of physical and functional rehabilitation, geriatrician, hematologist, psychiatrist, other medical staff, (masseur)-physiotherapist, speech therapist, nutritionist, ergo-therapist, psycho-motor therapist, assistant nurse, caregiver, nurse, specialized nurse, psychologist, educational staff, other non-medical staff, social worker) as well as equipment (isokinetic system, gait and movement analysis laboratory, equilibrium and posture exploration equipment, robotic walking assistance, robotic-assisted for rehabilitation plan, driving simulator, vehicle for person with reduced mobility, technical platform for urodynamic assessment, specific flat for dependents, room with living area, swimming-pool, balneotherapy, spine typology exploratory system, medical/surgical intervention ward for surgical dressings / u₄).

5. Case 5: Y^d=f((ownership, age group, Charlson Index, severity, age group crossed Charlson Index, age group crossed severity, the presence of chronic disease, provenance: nursing home, residence as well as entrance from the emergency unit, the type of specialization of the rehabilitation center, and rehabilitation center information, patient’s SES: deterioration in housing: unsanitary, not adapted to patient’s needs, no personal housing; homeless; living alone in her/his housing; isolated (no known family, no known relatives); the social coverage: inadequacy as judged by the medical staff during the acute care stay or the rehabilitation care stay; very low income, including patient covered by the National Health Insurance for the most economically disadvantaged; French EDI/ u4).

The comparison between Cases 4 and 5 addresses whether SES factors explain efficiency and performance.

Results

Baseline results

We now turn to preliminary statistics of different outcomes for the databases on the disorder considered, stroke: 30,585 admissions to rehabilitation centers in 2018, COPD: 27,421 admissions, heart failure: 34,284 admissions, and total hip replacement: 33,205 admissions. The data includes public hospitals, non-profit hospitals, and for-profit ones. About two-thirds of stroke, heart failure, or COPD patients are admitted to a public sector rehabilitation hospital. About half of total hip replacement patients are treated in for-profit rehabilitation hospitals. The average patient is over 70 years old (70 years old for stroke and COPD, 83 years old for heart failure, and 75 years old for total hip replacement). According to the Charlson Index measurements computed on acute care stays, for-profit institution patients’ case mix is, on average, less severe than public service hospital patients’. However, the market is not segmented by severity: a substantial fraction of the private institutions’ patients have comorbidities.

The proportion of patients living in a deprived area is much lower among for-profit rehabilitation center patients than for public sector ones. As documented by Chard³⁷ and elsewhere, for-profit facilities’ patients are, on average, less deprived than patients whom public sector hospitals treat.

The average patient treated by a for-profit hospital has a lower probability of in-hospital mortality compared to those treated by a public hospital, whatever the pathology or procedure considered (for heart failure, the mortality rate is 4.7% among patients admitted to a for-profit institute, compared to 8.9% for those admitted to a public hospital). Home discharge patients who choose for-profit rehabilitation centers, have a higher probability of returning home following their rehabilitation stays on average. 4 out of 5 private hospital patients for stroke and heart failure return home, versus less than 70% for public providers.

Concerning the PAI index, private hospitals devoted more time, on average, to patient care than public hospitals for all observed disorders.

Econometric results

This paper assesses performance by combining the outcome quality indicators and the providers’ activity index (PAI): performance means better quality outcomes while providing less intense care (lower PAI).

We analyze each quality indicator index through five incremental regressions. Case 1: we controlled for the hospital’s ownership; Case 2: we added control for patient clinical components (the patient demographic variables, the severity level as measured during the acute care stay); Case 3: we added control for the specialization of the rehabilitation unit or institution that may impact the patient; Case 4: we added control for the hospital organization regarding staff composition that may affect patient care; Case 5: we added patient’s SES information. Indeed, the driver of efficiency may be the patient’s SES. Supplementary Digital Material 1, Supplementary Text File 1 presents all regression details. In the main text, we present Case 1, Case 4, and Case 5 (Table II, III).

Table II. —Case 1, Case 4 and Case 5: Stroke and COPD.

	Model	OLS model	OLS model	LOGIT model	LOGIT model	OLS model
		Coefficients	Coefficients	Coefficients	Coefficients	coefficients
		Physical score improvement	Cognitive score improvement	P(Death)	P(Return-to-home)	Provider Activity Index (PAI)
Stroke
Case 1	For-profit	0.593***	0.166***	-0.649***	0.0797***	7.869***
		(0.0391)	(0.0205)	(0.0953)	(0.0279)	(1.013)
	Non-profit	0.376***	0.0618***	-0.845***	0.168***	15.55***
		(0.0414)	(0.0218)	(0.111)	(0.0300)	(1.083)
	Public	Reference	Reference	Reference	Reference	Reference
Case 4	For-profit	0.166***	0.0678**	-0.391***	0.0871*	-9.445***
		(0.0479)	(0.0266)	(0.123)	(0.0452)	(1.107)
	Non-profit	0.0518	-0.0185	-0.217	-0.135***	-3.364***
		(0.0491)	(0.0273)	(0.133)	(0.0465)	(1.136)
	Public	Reference	Reference	Reference	Reference	Reference
Case 5	For-profit	0.115**	0.0479	-0.353**	0.0368	-9.408***
		(0.0560)	(0.0312)	(0.143)	(0.0527)	(1.296)
	Non-profit	0.0470	-0.0189	0.159	-0.116**	-3.452***
		(0.0492)	(0.0274)	(0.132)	(0.0463)	(1.138)
	Public	Reference	Reference	Reference	Reference	Reference
COPD
Case 1	For-profit	0.283***	0.115***	-0.832***	0.546***	11.27***
		(0.0285)	(0.0166)	(0.0957)	(0.0347)	(0.770)
	Non-profit	-0.0260	0.0258	-1.038***	0.533***	9.305***
		(0.0294)	(0.0171)	(0.107)	(0.0360)	(0.796)
	Public	Reference	Reference	Reference	Reference	Reference
Case 4	For-profit	0.262***	0.0453**	-0.0475	0.0580	-1.139
		(0.0360)	(0.0213)	(0.123)	(0.0517)	(0.749)
	Non-profit	-0.0442	-0.0495**	-0.0404	0.0399	-2.996***
		(0.0384)	(0.0227)	(0.137)	(0.0564)	(0.797)
	Public	Reference	Reference	Reference	Reference	Reference
Case 5	For-profit	0.254***	0.0685***	-0.0783	0.0537	-1.173
		(0.0430)	(0.0255)	(0.154)	(0.0632)	(0.893)
	Non-profit	-0.0432	-0.0497**	-0.175	0.0937*	-3.075***
		(0.0384)	(0.0227)	(0.140)	(0.0564)	(0.797)
	Public	Reference	Reference	Reference	Reference	Reference

In the results table, we use the standard notation for significance. ***significant at 1%.

Table III. —Case 1, Case 4 and Case 5: Heart failure and Total hip replacement.

	MODEL	OLS model	OLS model	LOGIT model	LOGIT model	OLS model
		Coefficients	Coefficients	Coefficients	Coefficients	Coefficients
		Physical score improvement	Cognitive score improvement	P(Death)	P(Return-to-home)	Provider Activity Index (PAI)
Heart failure
Case 1	For-profit	0.549***	0.224***	-0.676***	0.322***	6.727***
		(0.0283)	(0.0161)	(0.0516)	(0.0248)	(0.583)
	Non-profit	0.178***	0.0282	-0.607***	0.293***	-0.318
		(0.0356)	(0.0202)	(0.0662)	(0.0315)	(0.738)
	Public	Reference	Reference	Reference	Reference	Reference
Case 4	For-profit	0.292***	0.135***	-0.197***	0.199***	-0.0817
		(0.0342)	(0.0196)	(0.0645)	(0.0343)	(0.554)
	Non-profit	0.0306	-0.0233	-0.349***	0.196***	-1.794***
		(0.0381)	(0.0217)	(0.0750)	(0.0389)	(0.616)
	Public	Reference	Reference	Reference	Reference	Reference
Case 5	For-profit	0.258***	0.149***	-0.0451	0.147***	-0.996
		(0.0428)	(0.0244)	(0.0802)	(0.0430)	(0.691)
	Non-profit	0.0273	-0.0210	-0.366***	0.216***	-1.672***
		(0.0382)	(0.0218)	(0.0751)	(0.0385)	(0.617)
	Public	Reference	Reference	Reference	Reference	Reference
Total hip replacement
Case 1	For-profit	0.357***	0.115***	-0.385	0.188***	2.328***
		(0.0302)	(0.0122)	(0.377)	(0.0423)	(0.687)
	Non-profit	0.266***	0.0881***	-0.629	0.0997**	-5.958***
		(0.0357)	(0.0144)	(0.500)	(0.0497)	(0.812)
	Public	Reference	Reference	Reference	Reference	Reference
Case 4	For-profit	0.171***	0.0964***	-0.123	0.301***	0.125
		(0.0321)	(0.0135)	(0.432)	(0.0606)	(0.508)
	Non-profit	0.0928**	0.0749***	-0.519	0.0463	-1.632***
		(0.0369)	(0.0156)	(0.570)	(0.0698)	(0.584)
	Public	Reference	Reference	Reference	Reference	Reference
Case 5	For-profit	0.171***	0.0955***	-0.0895	0.109*	-0.231
		(0.0333)	(0.0141)	(0.477)	(0.0631)	(0.527)
	Non-Profit	0.0912**	0.0749***	-0.633	0.0252	-1.753***
		(0.0371)	(0.0157)	(0.599)	(0.0688)	(0.588)
	Public	Reference	Reference	Reference	Reference	Reference

In the results table, we use the standard notation for significance. ***significant at 1%.

For stroke inpatients, compared to public hospitals, private hospitals (for-profit and non-profit) have better outcome indices of dependency improvement, higher probability of returning home, and lower probability of death (Case 1). According to Case 4 and even Case 5, outcome quality indicators remain significantly explained by ownership, but the values of the coefficients are much lower. Patient clinical characteristics, medical unit specialization of rehabilitation centers, the medical staff and equipment in the rehabilitation center, and the patient’s SES explain the efficiency.

According to the outcome quality index, some comparable performance is found between public and for-profit rehabilitation centers when adding SES factors. In addition, this better efficiency is obtained with minor PAI. Therefore, for-profit rehabilitation centers seem to perform better than other rehabilitation facilities.

We now turn to non-profit hospitals. Adding the SES factors further underscores the ambiguous results. There is no clear-cut difference in performance between non-profit and public hospitals (Case 5).

For COPD inpatients, comparing Case 1 to Case 4 reveals that the hospitals’ characteristics in terms of equipment and staff (medical and paramedical) explain large parts of the difference in ownership’s quality indicators. It leads to for-profit centers’ performance being closer to public rehabilitation services but nevertheless performs better (Case 4). These results are obtained with less care provided to patients, but not significantly less (at the 5% level). Adding the SES factors does not modify this result (Case 5).

From Case 1, the results of quality indicators’ differences between non-profit and public hospitals are not straightforward. Controlling patients’ clinical and demographic components and hospital characteristics (Case 4), non-profit rehabilitation centers do not perform better than public hospitals.

Whatever the hospital’s type, the values of the coefficients for ownership decrease strongly from Case 1 to Case 5, except for physical score improvement. Patient clinical characteristics, medical unit specialization of rehabilitation centers, the medical staff and equipment in the rehabilitation center, and the patient’s SES explain the efficiency.

For heart failure inpatients, we found comparable results to what we found for stroke patients. However, contrary to the effects of stroke patients, this result is obtained with no difference in PAI. Adding the SES factors emphasizes this result (Case 5).

Non-profit hospitals show quality indicators better than or comparable to public rehabilitation units, whatever the outcome. However, the coefficient for cognitive score improvement exhibits a negative sign that jeopardizes the conclusion on efficiency.

For total hip replacement, the expected low mortality rate for total hip replacement inpatients explains the absence of a link between the probability of mortality and the hospital’s ownership, whatever the model. Private for-profit rehabilitation hospitals provide better quality indicators than public rehabilitation services. When we added the SES factors, the results remained stable. In addition, we found this result with no significant difference in care activity for the average patient. For-profit hospitals seem to be the most efficient for this elective disorder.

Overall results reveal some common features of these four pathologies and acts. Patient clinical characteristics, medical unit specialization of rehabilitation centers, the medical staff and equipment in the rehabilitation center, and the patient’s SES explain the efficiency. Regarding the outcome quality index, we show no clear-cut difference between non-profit rehabilitation centers and public rehabilitation services. For-profit rehabilitation centers seem to outperform public ones. It may also be due to a patient sorting effect, i.e., the disparities in SES’s composition depending on the hospital’s ownership.

Counterfactual simulations: patient sorting effect in patients’ social SES

For-profit rehabilitation centers have emerged as more efficient than public ones. This result raises questions about whether gains in access to for-profit providers benefit patients equally and whether existing inequalities are attenuated or exacerbated. Indeed, the difference in quality indicators may be due to the patients’ admitted characteristics regarding clinical and demographic or SES composition. Private hospitals may admit patients with better underlying health than public hospitals. More precisely, private hospitals may admit patients with a higher potential for improvement during a rehabilitation stay than public rehabilitation centers. The selection process allows such discrimination where for-profit hospitals can select patients while public sector hospitals cannot. Counterfactual simulation analysis permits some indicative assessment of these questions (Supplementary Text File 1).

We run a linear model for patients treated in public hospitals. We then use this model to predict the outcome quality index for all patients, whatever the provider type. Our identification strategy suggests that unobserved factors are not being controlled for in each specification that are introducing bias. We use Case 5 for valid counterfactuals, whose all observed factors are controlled. We compute simulation for each quality indicator and each database according to the disorder selected (stroke, COPD, heart failure, and total hip replacement).

To illustrate our findings, we comment on physical dependence improvement results (Table IV).

Table IV. —Simulation results compared to the prediction results.

	Stroke				COPD
	Prediction+		Simulation&	Difference	Prediction		Simulation	Difference
	Public sector	Private sector	Private sector	% explained by SES	Public sector	Private sector	Private sector	% explained by SES
Physical score improvement	1.24	1.87	1.64	36.51%	0.30	0.58	0.50	28.22%
Cognitive score improvement	0.21	0.38	0.31	42.94%	-0.02	0.09	0.07	19.54%
P (death)	0.03	0.02	0.02	10.10%	0.04	0.02	0.02	9.93%
P (return-to-home)	0.61	0.64	0.62	44.28%	0.69	0.80	0.75	46.63%
% of obs. from public sector acute care unit	91.07%				83.06%

	Heart failure				Total hip replacement
	Prediction		Simulation	Difference	Prediction		Simulation	Difference
	Public sector	Private sector	Private sector	% explained by SES	Public sector	Private sector	Private sector	% explained by SES
Physical score improvement	0.33	0.91	0.79	21.05%	1.26	1.78	1.52	35.61%
Cognitive score improvement	-0.09	0.13	0.10	17.85%	0.07	0.18	0.17	17.31%
P (death)	0.10	0.05	0.07	45.37%	0.00	0.00	Not relevant	Not relevant
P (return-to-home)	0.56	0.64	0.63	16.91%	0.88	0.88	0.88	11.02%
% of obs. from public sector acute care unit	83.61%				48.56%

⁺ The predicted outcome quality indices using the models controlling for all set of variables including the SES index (Case 5). ^& We simulate the predicted outcome quality indices based on the population of patient j admitted to public rehabilitation centers k as if they had been admitted in a private rehabilitation center h.

The predicted value shows a 1.24 improvement in physical dependence for patients treated by public rehabilitation hospitals versus 1.87 for patients treated by for-profit ones (following a public sector acute care unit admission). We simulate as described above to investigate the importance of the patient’s sorting. The simulated value is 1.64 for patients treated by for-profit centers. (1.87-1.24=0.63 and 1.64-1.24=0.40, so 0.23/0.63=36.5%). These figures show that patients’ sorting in for-profit centers explains 36.5% of the performance difference. For-profit patients are less complex. Whatever the disorder and the quality indicators selected, one observes comparable results. The share due to SES in performance difference ranges from 10% to 45%.

Discussion

Sensitivity analysis

Private units or for-profit institutions comprise hospitals that can select their patients, while public sector hospitals cannot. Are the results obtained due to the selection of patients? In cases 2 to 5, we controlled using patient clinical variables, but it might not be sufficient to control all aspects of the patient’s case mix. Beyond the gap between the empirical clinical factors and the classification refinement translated into codes, differences observed in patients’ case mix may be due to differences in coding practices between ownership. Milcent³⁸ shows that optimizing coding behavior is performed more often in for-profit hospitals. We used only information from patients admitted to public sector acute care units to deal with this limitation. This patient group allows us to avoid coding differences from acute care stays that may affect the results. The results are comparable. As an additional robustness test, we run models whose severity coded in the rehabilitation centers is not considered. The main results are unchanged.

As a sensitivity analysis, we run the described counterfactual models for each sample on patients from a public acute care facility. Because of lack of public rehabilitation centers, there are agreements between facilities to admit patients from public acute care facilities. This is free of charge for the patient due to the national health insurance scheme. The results are unchanged.

As a sensitivity analysis, six samples were set up on the frequency of Major Diagnostic Category (MDC) stays from twenty-six MDC codes, e.g. 1) nervous system disorders; 2) respiratory disorders; 3) disorders of the circulatory system; 4) disorders or impairment of musculoskeletal system and connective tissue condition; 5) endocrine, metabolic and nutritional disorders; 6) disorders and mental illnesses. Whatever the affliction, the principal results remain the same.

Public hospitals comprise large public research hospitals and other small public hospitals. After splitting the public rehabilitation centers (public research hospitals and other public hospitals) into two groups, we find heterogeneity in performance between public research hospitals and other public hospitals. No conclusive results have been found. Results on for-profit hospitals remain unchanged.

Limitations of the study

In this work, individual SES is approximated by ICD-10 codes. This SES, therefore, relates to cases of extreme situations. Likely, differences in care are also observed for patients in less fragile situations. It is, therefore, necessary to continue this work with a specific survey on the economic and social conditions and, therefore, the degree of isolation of patients. This would allow a more detailed study of the effects of selecting patients for care on the performance of healthcare institutions.

Moreover, SES is multidimensional. The SES indicator studied here is an aggregate of several dimensions. To isolate the differentiated effects of social, economic, and childhood-related effects, it is appropriate to continue this study by differentiating these reasons for SES.

Conclusions

This article raises the question of the efficiency of the hospital when there is a sorting effect of the patient’s SES beyond the clinical components. To address this, we first explored how the clinical and demographic components impact the hospital’s efficiency. As a second step, we proposed a counterfactual simulation to investigate the role of sorting patients in the efficiency of rehabilitation services.

We used an exhaustive database for rehabilitation stays with admission in the year 2018 for the entire French population. Four indicators of quality in outcomes are used. The multiplication of outcomes allows accounting for multidimensionality and reinforcing the results. This study is carried out on four disorders: stroke, COPD, heart failure, and total hip replacement. We use models _OLS and Logistic regression_ in which dimensions were progressively added: Hospital ownership (Case 1), patient case-mix variables (Case 2), specialization of the rehabilitation unit (Case 3), rehabilitation centers’ staff composition (Case 4), and patient SES (Case 5). In doing so, we were able to identify the ownership’s difference in efficiency before and after controlling for these components.

It shows that these components impact efficiency, including the patient’s SES. Then, for-profit rehabilitation units seem more efficient than public sector hospitals. Turning to non-profit hospitals, no clear-cut results are found. As a sensitivity test, we added an analysis of six major diagnostics where similar results were found.

Using counterfactual simulations, we quantitatively assess the role of unobserved drivers of patient SES, explaining part of for-profit efficiency. The share due to SES in difference in performance is from 10% to 45%.

Despite almost free healthcare access, results demonstrate existing inequalities in healthcare access. The inequality of access to care is driven by differences in admission policy by for-profit centers. The conclusion drawn by this paper raises concerns about the way performance benefits are distributed across different types of patients. The performance of healthcare institutions is correlated not only to the case mix of their patients but also to the SES of the patients admitted. The performance needs to be seen in light of the SES of patients.

Supplementary Digital Material 1

Supplementary Text File 1

Method for the counterfactual