Public-private settlement and hospital mortality per sources of payment

Juliana Pires Machado Mônica Martins Iuri da Costa Leite About the authors

ABSTRACT

OBJECTIVE

To analyze if the adjusted hospital mortality varies according to source of payment of hospital admissions, legal nature, and financing settlement of hospitals.

METHODS

Cros-ssectional study with information source in administrative databases. Specific hospital admission reasons were selected considering the volume of hospital admissions and the list of quality indicators proposed by the North-American Agency for Healthcare Research and Quality (AHRQ). Were analyzed 852,864 hospital admissions of adults, occurred in 789 hospitals between 2008 and 2010, in Sao Paulo and Rio Grande do Sul, applying multilevel logistic regression.

RESULTS

At hospital admission level, showed higher chances of death male patients in more advanced age groups, with comorbidity, who used intensive care unit, and had the Brazilian Unified Health System as source of payment. At the level of hospitals, in those located in the mean of the distribution, the adjusted probability of death in hospital admissions financed by plan or private was 5.0%, against 9.0% when reimbursed by the Brazilian Unified Health System. This probability increased in hospital admissions financed by the Brazilian Unified Health System in hospitals to two standard deviations above the mean, reaching 29.0%.

CONCLUSIONS

In addition to structural characteristics of the hospitals and the profile of the patients, interventions aimed at improving care should also consider the coverage of the population by health plans, the network shared between beneficiaries of plans and users of the Brazilian Unified Health System, the standard of care to the various sources of payment by hospitals and, most importantly, how these factors influence the clinical performance.

Hospital Mortality; Private Health Care Coverage; Public-Private Sector Partnerships; Unified Health System; Outcome and Process Assessment (Health Care)

INTRODUCTION

The Brazilian health system presents peculiarities related to the way the public-private settlement results in health care. Hospital care is mainly carried out by private entities, which can simultaneously attended patients funded by the Brazilian Unified Health System (SUS) and by private health plans, producing complex financing settlements and assistance networks that make the regulation of the system difficult.

The interaction between SUS and health plans has been considered a factor that produces inequality in the access, use and, quality of health services2222. Santos IS. Evidência sobre o mix público-privado em países com cobertura duplicada: agravamento das iniquidades e da segmentação em sistemas nacionais de saúde. Cienc Saude Coletiva. 2011;16(6):2743-52. DOI:10.1590/S1413-81232011000600013
https://doi.org/10.1590/S1413-8123201100...
. In addition, the public-private competition for the same network of providers, without separation of the clientele served, raises problems in the organization of care and impairs the relationship of complementarity expected for the system22. Basu S, Andrews J, Kishore S, Panjabi R, Stuckler D. Comparative performance of private and public healthcare systems in low and middle-income countries: a systematic review. PLoS Med. 2012;9(6):E100124. DOI:10.1371/journal.pmed.1001244
https://doi.org/10.1371/journal.pmed.100...
.

Hospital mortality has been one of the most exploited indicators to measure hospital quality for representing an unequivocal result, a global measure of the care process, and also for its availability in the administrative databases55. Burnett S, Renz A, Wiig S, Fernandes A, Weggelaar AM, Calltorp J et al. Prospects for comparing European hospitals in terms of quality and safety: lessons from a comparative study in five countries. Int J Qual Health Care. 2013;25(1):1-7. DOI:10.1093/intqhc/mzs079
https://doi.org/10.1093/intqhc/mzs079...
,66. Campbell MJ, Jacques RM, Fotheringham J, Maheswaran R, Nicholl J. Developing a summary hospital mortality index: retrospective analysis in English hospitals over five years. BMJ. 2012;344:e1001. DOI:10.1136/bmj.e1001
https://doi.org/10.1136/bmj.e1001...
,1919. Pitches DW, Mohammed MA, Lilford RJ. What is the empirical evidence that hospitals with higher-risk adjusted mortality rates provide poorer quality care? A systematic review of the literature. BMC Health Serv Res. 2007;7(1):91. DOI:10.1186/1472-6963-7-91
https://doi.org/10.1186/1472-6963-7-91...
. Hospital mortality must be adjusted for the risk of the patients, to enable more reliable comparisons of providers. Traditionally, these adjustments include, in regression models, attributes of patients expressing degrees of risk of death66. Campbell MJ, Jacques RM, Fotheringham J, Maheswaran R, Nicholl J. Developing a summary hospital mortality index: retrospective analysis in English hospitals over five years. BMJ. 2012;344:e1001. DOI:10.1136/bmj.e1001
https://doi.org/10.1136/bmj.e1001...
,1313. Iezzoni LI. Risk adjustment for performance measurement. In: Smith PC, Mossialos E, Papanicolas I, Leatherman S. Performance measurement for health system improvement: experiences, challenges and prospects. Cambridge: Cambridge University Press; 2009. p. 251-368.,1616. Martins M. Uso de medidas de comorbidades para predição de risco de óbito em pacientes brasileiros hospitalizados. Rev Saude Publica. 2010;44(3):448-56 . DOI:10.1590/S0034-89102010005000003
https://doi.org/10.1590/S0034-8910201000...
,2323. Sharabiani MTA, Aylin P, Bottle A. Systematic review of comorbidity indices for administrative data. Med Care. 2012;50(12):1109-18. DOI:10.1097/MLR.0b013e31825f64d0
https://doi.org/10.1097/MLR.0b013e31825f...
. In addition to the characteristics of the patients, the structure and functioning of hospitals may also affect hospital mortality. Therefore, the variation in hospital mortality can indicate problems in the quality of the care provided by hospitals. In this sense, to measure different explanatory factors involved is crucial in this approach, although it is not trivial.

In addition, complicating this type of analysis, there is the fact that patients in the same hospital have more similarity between themselves than with patients in different hospitals, generating dependency between observations within each hospital. To deal with this dependency, multilevel regression models have been used in some studies2121. Sanagou M, Wolfe R, Forbes A, Reid CM. Hospital-level associations with 30-day patient mortality after cardiac surgery: a tutorial on the application and interpretation of marginal and multilevel logistic regression. BMC Med Res Methodol. 2012;12(1):28. DOI:10.1186/1471-2288-12-28
https://doi.org/10.1186/1471-2288-12-28...
.

The relationship between hospital mortality and structural and organizational characteristics of hospitals has received great attention in national and international literature44. Brand CA, Barker AL, Morello RT, Vitale MR, Evans SM, Scott IA et al. A review of hospital characteristics associated with improved performance. Int J Qual Health Care. 2012;24(5):483-94. DOI:10.1093/intqhc/mzs044
https://doi.org/10.1093/intqhc/mzs044...
, with emphasis on the analysis of public or private nature22. Basu S, Andrews J, Kishore S, Panjabi R, Stuckler D. Comparative performance of private and public healthcare systems in low and middle-income countries: a systematic review. PLoS Med. 2012;9(6):E100124. DOI:10.1371/journal.pmed.1001244
https://doi.org/10.1371/journal.pmed.100...
,88. Eggleston K, Shen YC, Lau J, Schmid CH, Chan J. Hospital ownership and quality of care: what explains the different results in the literature? Health Econ. 2008;17(12):1345-62. DOI:10.1002/hec.1333
https://doi.org/10.1002/hec.1333...
and on source of financing for hospital admissions2525. Spencer CS, Gaskin DJ, Roberts ET. The quality of care delivered to patients within the same hospital varies by insurance type. Health Aff (Millwoood). 2013;32(10):1731-9. DOI:10.1377/hlthaff.2012.1400
https://doi.org/10.1377/hlthaff.2012.140...
,2727. Weissman JS, Vogeli C, Levy DE. The quality of hospital care for Medicaid and private pay patients. Med Care. 2013;51(5):389-95. DOI:10.1097/MLR.0b013e31827fef95
https://doi.org/10.1097/MLR.0b013e31827f...
. In Brazil, studies developed, in general, are restricted to certain geographical areas1212. Gomes AS, Klück MM, Fachel JMG, Riboldi J. Fatores associados à mortalidade hospitalar na rede SUS do Rio Grande do Sul, em 2005: aplicação de modelo multinível. Rev Bras Epidemiol. 2010;13(3):533-42. DOI:10.1590/S1415-790X2010000300016
https://doi.org/10.1590/S1415-790X201000...
,1515. Martins M, Blais R, Leite IC. Mortalidade hospitalar e tempo de permanência: comparação entre hospitais públicos e privados na região de Ribeirão Preto, São Paulo, Brasil. Cad Saude Publica. 2004;20 Suppl 2:S268-82. DOI:10.1590/S0102-311X2004000800021
https://doi.org/10.1590/S0102-311X200400...
. This is mainly due to the unavailability of secondary data with acceptable coverage and quality of fulfillment to the whole Brazilian territoryaaMachado JP. O arranjo público-privado no Brasil e a qualidade da assistência hospitalar em São Paulo e no Rio Grande do Sul [these]. Rio de Janeiro (RJ): Escola Nacional de Saúde Pública, Fundação Oswaldo Cruz; 2014..

This study aimed to analyze if adjusted hospital mortality varies according to source of payment of hospital admissions, legal nature, and financing settlement of hospitals.

METHODS

Study universe

Were studied hospital admissions carried out between 2008 and 2010 in hospitals in Sao Paulo and Rio Grande do Sul, with better coverage and quality of fulfillment of Hospital Admissions Communication (CIH)aaMachado JP. O arranjo público-privado no Brasil e a qualidade da assistência hospitalar em São Paulo e no Rio Grande do Sul [these]. Rio de Janeiro (RJ): Escola Nacional de Saúde Pública, Fundação Oswaldo Cruz; 2014..

Of 39,419,539 hospital admissions recorded during this period, 7,385,323 were selected for patients aged between 18 and 99 years, admitted less than 30 days (acute), not transferred, in the specialties general surgery or internal medicine, whose procedures and diagnoses were unrelated to obstetrics, not occurring in hospitals with at least one hospital admission/day.

From these 7,385,323 hospital admissions, the study universe integrated the 852,864 regarding four causes with more admissions and deaths in Brazil, among the seven proposals for quality indicators for the Agency for Healthcare Research and Quality – AHRQ)bbDepartment of Health and Human Services (UEA), Agency for Healthcare Research and Quality. Guide to inpatient quality indicators: quality of care in hospitals: volume, mortality, and utilization. Version 3.1, Rockville, MD: Agency for Healthcare Research and Quality; 2007.. They are: acute myocardial infarction – AMI (CID 21); congestive heart failure – CHF (CID I50, I11, I13); cerebrovascular accident – CVA (CID I60 I61 I62 CID, I63, I64); and pneumonia (CID J13, J14, J15, J16, J18).

The use of these causes is based on greater causal validity of hospital mortality because of the care applied to such diagnoses, with evidences indicating relationship between process of care and survival of the patientsbbDepartment of Health and Human Services (UEA), Agency for Healthcare Research and Quality. Guide to inpatient quality indicators: quality of care in hospitals: volume, mortality, and utilization. Version 3.1, Rockville, MD: Agency for Healthcare Research and Quality; 2007.. The same causes are used in monitoring and evaluations of quality in other countries, including comparisons between them55. Burnett S, Renz A, Wiig S, Fernandes A, Weggelaar AM, Calltorp J et al. Prospects for comparing European hospitals in terms of quality and safety: lessons from a comparative study in five countries. Int J Qual Health Care. 2013;25(1):1-7. DOI:10.1093/intqhc/mzs079
https://doi.org/10.1093/intqhc/mzs079...
,1919. Pitches DW, Mohammed MA, Lilford RJ. What is the empirical evidence that hospitals with higher-risk adjusted mortality rates provide poorer quality care? A systematic review of the literature. BMC Health Serv Res. 2007;7(1):91. DOI:10.1186/1472-6963-7-91
https://doi.org/10.1186/1472-6963-7-91...
.

Databases

Were used administrative databases provided by the Department of Informatics of SUS (DATASUS) and by the National Health Agency (ANS), both linked to the Brazilian Ministry of Health.

For the SUS-financed admissions, data are from the System of Hospital Information of SUS (SIH/DATASUS); for the admissions whose source of payment were private or of health plan, data have as a source the CIH/DATASUS. For both cases are publicly available on the DATASUS website files with individualized data (however de-identified) of patients.

Data on the structure of public and private hospital network came from the National Registry of Health Establishments (CNES/DATASUS), also obtained in the website. Further data on private network have been obtained from the System of Health Plans Registry (RPS/ANS), upon request of information to the ANS.

Analyses

The dependent variable was the occurrence of in-hospital death. Data analysis was carried out in two steps: (1) risk adjustment model; (2) analysis of characteristics of the care process and the structure of hospitals.

The risk adjustment model included the following variables: sex, age, primary diagnosis, presence of comorbidity, Charlson Comorbidity Index (CCI), and clinical conditions of Elixhauser not covered by CCI. Several models were tested from the base model composed by age and sex, then including other variables. The variable type of admission (elective or emergency) has not been tested because it was not fulfilled in the CIH database. Although information about comorbidity are restricted in the sources of information used, the two comorbidity indexes have been used aiming to greater precision of risk adjustment models, qualifying them to represent a degree of information about comorbidity. The algorithm of Quan et al.1616. Martins M. Uso de medidas de comorbidades para predição de risco de óbito em pacientes brasileiros hospitalizados. Rev Saude Publica. 2010;44(3):448-56 . DOI:10.1590/S0034-89102010005000003
https://doi.org/10.1590/S0034-8910201000...
was used in the construction of these indexes.

To assess the predictive ability of the models tested, the C-statistics (Receiver Operating Characteristics – ROC curve scores) was used. This test measures the probability predicted of risk of death in a randomly selected patient who died, compared to the probability of risk of death in a randomly selected patient who survived. For this statistics, the value of 0.5 suggests that the model is indifferent to a random chance of predicting death, while 1.0 suggests perfect discrimination; values up to 0.7 are considered of low discrimination, between 0.7 and 0.8, of moderate discrimination, and above 0.8, the model is considered predictive and of high discrimination11. Aylin P, Bottle A, Majeed A. Use of administrative data or clinical databases as predictors of risk of death in hospital: comparison of models. BMJ Online. 2007;334:1044. DOI:10.1136/bmj.39168.496366.55
https://doi.org/10.1136/bmj.39168.496366...
.

In the second step, variables regarding hospital admission level were: length of stay, type of procedure performed (surgical or clinical), use of Intensive Care Unit (ICU), and source of payment of hospital admission (SUS, health plan, or private). The variables regarding hospital level included were: legal nature, financing settlement, capacity, and educational activity.

As independent central variables in this study, it has been privileged: source of payment of the hospital admission, legal nature (public, for-profit private and non-profit private), and financing settlement (only SUS; SUS, health plans, and private; health plan and private) of the hospital.

The variable financing settlement is one of the types of institutional settlements, a term from the field of economics to describe forms of organization between market agents2020. Rolim CLRC, Martins M. Qualidade do cuidado ao acidente vascular cerebral isquêmico no SUS. Cad Saude Publica. 2011;27(11):2106-16. DOI:10.1590/S0102-311X2011001100004
https://doi.org/10.1590/S0102-311X201100...
. The financing settlement was built based on registration information and hospital production, each category indicating a combination in the sources of payment of admissions carried out in the hospital: “only SUS” refers to hospitals registered in the CNES with agreement only to SUS, not listed in the RPS, and no admissions in CIH; “SUS, health plans, and private” refers to those registered in the CNES with agreements with SUS, health plans, or private and included or not in the RPS, with admissions in SIH and CIH, or those registered in the CNES with agreement only with SUS, however included in the RPS or admissions in CIH; and “health plan and private” refers to those registered in the CNES with health plans or private agreements, included or not in the RPS, with no admissions in SIH.

Multilevel Logistic Regression Model

The multilevel logistic regression model with two levels (patient and hospital) used2121. Sanagou M, Wolfe R, Forbes A, Reid CM. Hospital-level associations with 30-day patient mortality after cardiac surgery: a tutorial on the application and interpretation of marginal and multilevel logistic regression. BMC Med Res Methodol. 2012;12(1):28. DOI:10.1186/1471-2288-12-28
https://doi.org/10.1186/1471-2288-12-28...
is expressed as follows:

Where: pij is the probability of the ith patient of the jth hospital dying during admission; ß is the vector of parameters associated with the matrix of covariates of xij patients; γ is the vector of parameters associated with the variables of the hospital, that is, the second level of the multilevel model; u0j is the random parameter that measures the variability of outcome between hospital units. It is assumed that u0j has Normal distribution with zero mean and σ2ou variance.

The parameters of multilevel models were estimated using MLwiN software, version 2.32. We used the Predictive near-likelihood procedure of second-order, considering the estimation process of MLwiN most suitable for multilevel models of binary response1111. Goldstein H, Rasbash J. Improved approximations for multilevel models with binary responses. JR Stat Soc. 1996;59(3):505-13. DOI:10.2307/2983328
https://doi.org/10.2307/2983328...
. While standard deviation of the random effect can be directly estimated by the square root of the variance, the calculation of its standard error was obtained by Delta method77. Casella GE, Berger RL. Statistical inference. 2nd ed. Pacific Grove: Duxbury; 2002..

The effects of variables regarding patients and hospitals were interpreted in terms of ratio of chance and their respective 95% confidence intervals. The residual effect of the hospitals was estimated at probabilities of death, calculated according to different levels of random effect: -2 standard deviations below mean, -1 standard deviation below mean, at mean, +1 standard deviation above mean, and +2 standard deviations above mean.

RESULTS

Of 852,864 hospital admissions selected, 41.0% were due to pneumonia, 30.0% to CHF, 19.0% to CVA, and 10.0% to AMI. The gross mortality rate was 13.5%, with variations of 10.3% (CHF) and 17.3% (CVA) between the selected causes (Table 1).

Table 1
Characteristics of the study population. Sao Paulo and Rio Grande do Sul, Brazil, 2008-2010.

Most admissions were of patients between 60 and 79 years of age. The minority of them possessed some comorbidity, Charlson index other than zero, or presented some Elixhauser comorbidity. Most admissions were for up to seven days. Mostly, the ICU was not used and surgeries were not performed. The payment of hospital admissions was predominantly SUS. Admissions occurred mostly in private hospitals, without teaching activity, capacity greater than 49 beds, and mixed financing settlement: SUS, health plans, and private (Table 1). In hospitals with mixed settlement, admissions paid by SUS were majority. In hospitals of health plans and private settlement, admissions by health plans represented almost the total.

Regarding the risk adjustment model (Table 2), the better capacity of discrimination was from the model 10 (C-statistics = 0.66), which incorporated sex, age group, CCI, Elixhauser comorbidities excluded from CCI, presence of comorbidity, and main diagnosis (Table 2). In the multilevel model (Table 3), higher chances of death were registered in admissions of male patients, in more advanced age groups, with comorbidity, with Charlson index greater than zero, who remained for one day at the hospital, used ICU, performed clinical procedure, and with SUS as source of payment. At the level of hospitals, only the variable capacity proved to be statistically significant, with higher chance of death among larger hospitals (Table 3).

Table 2
Risk adjustment models tested. Sao Paulo and Rio Grande do Sul, Brazil, 2008-2010.
Table 3
Multilevel logistic regression model of hospital mortality: ratio of chances of death and estimated confidence intervals. Sao Paulo and Rio Grande do Sul, Brazil, 2008-2010.

The random effect, regarding the inter-hospital variation unexplained by the variables included in the model, was statistically significant, indicating that the chance of death of patients admitted to hospitals within one standard deviation above the mean e was 98.0% higher than that of patients admitted to hospitals on mean distribution. The Figure shows the probabilities of death for the categories of the variable source of payment according to random effect levels, that depict the variability in mortality between hospitals. Between hospitals within the mean, the greater probability of death was observed in patients with admissions paid by SUS, almost twice of those with health plan or private sources of payment. The probability of death increased significantly among patients admitted to hospitals within two standard deviations above the mean. In the case of patients with hospital admission paid by SUS, the probability of death reached 29.0%. Although admissions financed by health plan or private report less probability to death, this relationship can be changed depending on the hospital in which the patient was admitted to: admissions with health plan as source of payment, if occurred in hospitals within two standard deviations above the mean, had almost twice the probability of admissions whose source of payment was SUS, when admitted to hospitals within the mean (Figure).

Figure
Probability of death per source of payment of hospital admission, according to values of random effects estimated in the multilevel model. Sao Paulo and Rio Grande do Sul, Brazil, 2008-2010.

DISCUSSION

In Brazil, few studies analyze the quality of hospital care measured by risk-adjusted mortality employing multilevel models. Yet, the results found align to those of national studies identified1616. Martins M. Uso de medidas de comorbidades para predição de risco de óbito em pacientes brasileiros hospitalizados. Rev Saude Publica. 2010;44(3):448-56 . DOI:10.1590/S0034-89102010005000003
https://doi.org/10.1590/S0034-8910201000...
,2020. Rolim CLRC, Martins M. Qualidade do cuidado ao acidente vascular cerebral isquêmico no SUS. Cad Saude Publica. 2011;27(11):2106-16. DOI:10.1590/S0102-311X2011001100004
https://doi.org/10.1590/S0102-311X201100...
. At the international level, if on one hand the selection of four of the seven groups of causes recommended by AHRQ brings the need for adaptations for comparison between publications with similar approaches, on the other hand, there is variability in the set of selected code in European studies, corroborating the need of local adaptations as the one carried out55. Burnett S, Renz A, Wiig S, Fernandes A, Weggelaar AM, Calltorp J et al. Prospects for comparing European hospitals in terms of quality and safety: lessons from a comparative study in five countries. Int J Qual Health Care. 2013;25(1):1-7. DOI:10.1093/intqhc/mzs079
https://doi.org/10.1093/intqhc/mzs079...
,1919. Pitches DW, Mohammed MA, Lilford RJ. What is the empirical evidence that hospitals with higher-risk adjusted mortality rates provide poorer quality care? A systematic review of the literature. BMC Health Serv Res. 2007;7(1):91. DOI:10.1186/1472-6963-7-91
https://doi.org/10.1186/1472-6963-7-91...
.

Regarding risk adjustment, the variables here used have similarities with those used in international and national studies, for their recognized relationship with the risk of death and availability in databases11. Aylin P, Bottle A, Majeed A. Use of administrative data or clinical databases as predictors of risk of death in hospital: comparison of models. BMJ Online. 2007;334:1044. DOI:10.1136/bmj.39168.496366.55
https://doi.org/10.1136/bmj.39168.496366...
,66. Campbell MJ, Jacques RM, Fotheringham J, Maheswaran R, Nicholl J. Developing a summary hospital mortality index: retrospective analysis in English hospitals over five years. BMJ. 2012;344:e1001. DOI:10.1136/bmj.e1001
https://doi.org/10.1136/bmj.e1001...
,1212. Gomes AS, Klück MM, Fachel JMG, Riboldi J. Fatores associados à mortalidade hospitalar na rede SUS do Rio Grande do Sul, em 2005: aplicação de modelo multinível. Rev Bras Epidemiol. 2010;13(3):533-42. DOI:10.1590/S1415-790X2010000300016
https://doi.org/10.1590/S1415-790X201000...
,2323. Sharabiani MTA, Aylin P, Bottle A. Systematic review of comorbidity indices for administrative data. Med Care. 2012;50(12):1109-18. DOI:10.1097/MLR.0b013e31825f64d0
https://doi.org/10.1097/MLR.0b013e31825f...
,2525. Spencer CS, Gaskin DJ, Roberts ET. The quality of care delivered to patients within the same hospital varies by insurance type. Health Aff (Millwoood). 2013;32(10):1731-9. DOI:10.1377/hlthaff.2012.1400
https://doi.org/10.1377/hlthaff.2012.140...
,2727. Weissman JS, Vogeli C, Levy DE. The quality of hospital care for Medicaid and private pay patients. Med Care. 2013;51(5):389-95. DOI:10.1097/MLR.0b013e31827fef95
https://doi.org/10.1097/MLR.0b013e31827f...
. On the other hand, in relation to information about comorbidity and related indexes, although less used in Brazilian studies because of the incompleteness of the records1616. Martins M. Uso de medidas de comorbidades para predição de risco de óbito em pacientes brasileiros hospitalizados. Rev Saude Publica. 2010;44(3):448-56 . DOI:10.1590/S0034-89102010005000003
https://doi.org/10.1590/S0034-8910201000...
, they were included in this study aiming at improving the predictive ability of risk adjustment model and, above all, the conceptual coherence with the analytical approach.

The treatment applied to length of stay and some of the results were similar to those studies that use global approach of hospital mortality1414. Jarman B, Pieter D, Veen AA, Kool RB, Aylin P, Bottle A et al. The hospital standardised mortality ratio: a powerful tool for Dutch hospitals to assess their quality of care? Qual Saf Health Care. 2010;(19):9-13. DOI:10.1136/qshc.2009.032953
https://doi.org/10.1136/qshc.2009.032953...
,1717. Mohammed MA, Deeks JJ, Girling A, Rudge G, Carmalt M, Stevens AJ et al. Evidence of methodological bias in hospital standardised mortality ratios: retrospective database study of English hospitals. BMJ. 2009;338:b780. DOI:10.1136/bmj.b780
https://doi.org/10.1136/bmj.b780...
. The high risk of death observed on the first day of admission is possibly related to emergency cases, especially those requiring palliative care or with lower therapeutic possibilities1717. Mohammed MA, Deeks JJ, Girling A, Rudge G, Carmalt M, Stevens AJ et al. Evidence of methodological bias in hospital standardised mortality ratios: retrospective database study of English hospitals. BMJ. 2009;338:b780. DOI:10.1136/bmj.b780
https://doi.org/10.1136/bmj.b780...
. On the other hand, the global approach of hospital mortality uses the length of stay as adjustment variable1414. Jarman B, Pieter D, Veen AA, Kool RB, Aylin P, Bottle A et al. The hospital standardised mortality ratio: a powerful tool for Dutch hospitals to assess their quality of care? Qual Saf Health Care. 2010;(19):9-13. DOI:10.1136/qshc.2009.032953
https://doi.org/10.1136/qshc.2009.032953...
, option here excluded, since the length of stay may express greater gravity, adverse events resulting from problems in the quality of care or availability of beds for long term care1515. Martins M, Blais R, Leite IC. Mortalidade hospitalar e tempo de permanência: comparação entre hospitais públicos e privados na região de Ribeirão Preto, São Paulo, Brasil. Cad Saude Publica. 2004;20 Suppl 2:S268-82. DOI:10.1590/S0102-311X2004000800021
https://doi.org/10.1590/S0102-311X200400...
.

Regarding the association between hospital mortality and hospital characteristics, we expected to find a lower risk of death in larger hospitals, for their better structure and for the relationship between volume and quality described in literature33. Birkmeyer JD, Dimick JB, Staiger DO. Operative mortality and procedure volume as predictors of subsequent hospital performance. Ann Surg. 2006;243(3):411-7. DOI:10.1097/01.sla.0000201800.45264.51
https://doi.org/10.1097/01.sla.000020180...
. However, the analysis showed a higher risk in larger hospitals, similar to what Garcia et al.1212. Gomes AS, Klück MM, Fachel JMG, Riboldi J. Fatores associados à mortalidade hospitalar na rede SUS do Rio Grande do Sul, em 2005: aplicação de modelo multinível. Rev Bras Epidemiol. 2010;13(3):533-42. DOI:10.1590/S1415-790X2010000300016
https://doi.org/10.1590/S1415-790X201000...
observed. In addition, the gradient of risk of death was similar between capacity classes, differentiating only hospitals with capacity smaller or greater than 50 beds. Despite the minimum cut of hospital volume in one case a day, smaller hospitals seem to present specific role with some degree of experience in hospital care of the causes studied, factor that may contribute to the better outcome of care.

The analysis showed the effect of the source of payment on the risk of death. Patients with private health plan or that pay out their own pocket (private) showed lower adjusted mortality rates than patients of SUS, although admitted to the same hospitals. Similar results were published in international studies, mainly from the USA, where the benefits are of patients covered by private insurances, when compared with those covered by public insurances2525. Spencer CS, Gaskin DJ, Roberts ET. The quality of care delivered to patients within the same hospital varies by insurance type. Health Aff (Millwoood). 2013;32(10):1731-9. DOI:10.1377/hlthaff.2012.1400
https://doi.org/10.1377/hlthaff.2012.140...
,2727. Weissman JS, Vogeli C, Levy DE. The quality of hospital care for Medicaid and private pay patients. Med Care. 2013;51(5):389-95. DOI:10.1097/MLR.0b013e31827fef95
https://doi.org/10.1097/MLR.0b013e31827f...
. In Brazil, Martins et al.1515. Martins M, Blais R, Leite IC. Mortalidade hospitalar e tempo de permanência: comparação entre hospitais públicos e privados na região de Ribeirão Preto, São Paulo, Brasil. Cad Saude Publica. 2004;20 Suppl 2:S268-82. DOI:10.1590/S0102-311X2004000800021
https://doi.org/10.1590/S0102-311X200400...
also found higher risk of death among patients of SUS, however, they did not studied possible sources of payment disparities within the same hospitals, here explored by the analysis of financing settlements.

Studies conducted in the United States showed higher risk of death among uninsured patients and with private payment88. Eggleston K, Shen YC, Lau J, Schmid CH, Chan J. Hospital ownership and quality of care: what explains the different results in the literature? Health Econ. 2008;17(12):1345-62. DOI:10.1002/hec.1333
https://doi.org/10.1002/hec.1333...
,2525. Spencer CS, Gaskin DJ, Roberts ET. The quality of care delivered to patients within the same hospital varies by insurance type. Health Aff (Millwoood). 2013;32(10):1731-9. DOI:10.1377/hlthaff.2012.1400
https://doi.org/10.1377/hlthaff.2012.140...
. This difference may be related to different eligibility of patients to public health services in both countries. In Brazil, the entire population is eligible to SUS; in the USA, some people are ineligible for public insurance and also do not have private insurance. Thus, while private patients in the USA are mostly excluded from both modalities (public and private), in Brazil patients who pay private hospitalization tend to have higher purchasing power or are beneficiaries of the best health plans, with good margins of reimbursement.

Of the factors that contribute to disparities in the risk of death among sources of payment we highlight the differences between clinical practice, access to technology, and procedures of high cost and complexity88. Eggleston K, Shen YC, Lau J, Schmid CH, Chan J. Hospital ownership and quality of care: what explains the different results in the literature? Health Econ. 2008;17(12):1345-62. DOI:10.1002/hec.1333
https://doi.org/10.1002/hec.1333...
. Analyzing the Brazilian health system, Victora et al.2626. Victora CG, Barreto ML, Leal MC, Monteiro CA, Schmidt MI, Paim J et al Health conditions and health-policy innovations in Brazil: the way forward. Lancet. 2011;377(9792):2042-53. DOI:10.1016/S0140-6736(11)60055-X
https://doi.org/10.1016/S0140-6736(11)60...
suggests that service providers that attend SUS patients and non-SUS patients offer differentiated standard of care according to the values they receive, which would influence the choice of procedure and material used, thus affecting quality of care and possibly the risk of death or occurrence of other adverse results.

Although there were privileged variables in this study, the effect of the financing settlement and legal nature of the hospital on the adjusted mortality showed no significance. However, although this feature had not been able to differentiate the risk of death in hospitals, we observe, within the same physical structures, differences between patients with source of payment SUS and non-SUS. This suggests that even physically available in hospitals, some resources are not available to patients of SUS, indicating iniquities in the process of hospital care.

Some limitations are inherent to the use of administrative databases, often designed originally to billing services. Therefore, their content may not comprise the whole set of information necessary for analyses of quality of services, or there may be incomplete or incorrect data collection, affecting the analyses developed. Nevertheless, the use of such data is an alternative explored in many countries due to the ease of obtaining, comprehensiveness, and continuity1515. Martins M, Blais R, Leite IC. Mortalidade hospitalar e tempo de permanência: comparação entre hospitais públicos e privados na região de Ribeirão Preto, São Paulo, Brasil. Cad Saude Publica. 2004;20 Suppl 2:S268-82. DOI:10.1590/S0102-311X2004000800021
https://doi.org/10.1590/S0102-311X200400...
,aaMachado JP. O arranjo público-privado no Brasil e a qualidade da assistência hospitalar em São Paulo e no Rio Grande do Sul [these]. Rio de Janeiro (RJ): Escola Nacional de Saúde Pública, Fundação Oswaldo Cruz; 2014.. This study highlights the use of CIH, which, despite representing the only source about non-SUS admissions in the country, is little used in Brazilian studies, often due to its questionable quality. Considering the variables of interest and the need to use the CIH to the proposed analyses, we opted for the delimitation of the study universe in two states with best coverage, in which we also observed reasonable degree of consistency in the data, according to a study of MachadoaaMachado JP. O arranjo público-privado no Brasil e a qualidade da assistência hospitalar em São Paulo e no Rio Grande do Sul [these]. Rio de Janeiro (RJ): Escola Nacional de Saúde Pública, Fundação Oswaldo Cruz; 2014..

Incompleteness of data on the severity profile of patients in Brazilian information systems, both public and private, outstands as another limitation. Incompleteness has been discussed in evaluation studies, since it prevents more accurate risk adjustments, compromising the accuracy of the analysis. In this sense, higher risk of death in public hospitals observed in this study may reflect failures in risk adjustment for not precisely measuring the greater severity of patients who use public services, possibly those with the worst socioeconomic conditions. However, the use of the source of payment as indicative of the health condition of the patient at the time of admission would not be an appropriate solution, since it would exclude from the analysis problems in the quality of care provided associated with discrimination or inequity. A specific restraint was the noninclusion of the risk model of the type of admission (elective or emergency), due to its nonfulfillment at the base of CIH, since national studies found relationship with the chance of death1212. Gomes AS, Klück MM, Fachel JMG, Riboldi J. Fatores associados à mortalidade hospitalar na rede SUS do Rio Grande do Sul, em 2005: aplicação de modelo multinível. Rev Bras Epidemiol. 2010;13(3):533-42. DOI:10.1590/S1415-790X2010000300016
https://doi.org/10.1590/S1415-790X201000...
,2020. Rolim CLRC, Martins M. Qualidade do cuidado ao acidente vascular cerebral isquêmico no SUS. Cad Saude Publica. 2011;27(11):2106-16. DOI:10.1590/S0102-311X2011001100004
https://doi.org/10.1590/S0102-311X201100...
.

The restricted degree of some categorical variables brought difficulties and also represents a limitation in the analyses. Especially in the case of financing settlements, each of the three categories studied includes groups of hospitals with varied proportion of sources of payment of admissions that compose its clientele. Are distinct hospitals that attend 20.0% or 80.0% of the patients by SUS, and this difference possibly influenced in varying degrees the result of the care, especially considering that sources of showed significant effect on the risk of death.

Although presenting limits that point to the need for improvement and expansion of clinical information registered in databases, the approach used in this study to assess hospital mortality, with the application of risk adjustment, multilevel regression risk, and use of administrative data, allowed more accurate estimation of the effect of the characteristics of admissions and hospitals on the risk of death. Even after controlling the effects of individual risk and considering the hierarchy between the levels of analysis, the findings of this study indicate the existence of differences in the quality of hospital care, measured by adjusted mortality, according to sources of payment of the hospital admission. This analysis indicated a disadvantage for SUS patients when compared with patients of health plans or with private payment, including when admitted to the same hospital. In addition to structural characteristics of the hospitals and the profile of the patients, the elaboration of policies aimed at this area should also consider the coverage of the population by health plans, the network shared between beneficiaries of plans and users of SUS, the dynamics of attendance to the various sources of payment by hospitals and, most importantly, how these factors influence the clinical performance. Therefore, they would contribute to the reduction of inequalities and to improve the effectiveness of the health system.

REFERENCES

  • 1
    Aylin P, Bottle A, Majeed A. Use of administrative data or clinical databases as predictors of risk of death in hospital: comparison of models. BMJ Online. 2007;334:1044. DOI:10.1136/bmj.39168.496366.55
    » https://doi.org/10.1136/bmj.39168.496366.55
  • 2
    Basu S, Andrews J, Kishore S, Panjabi R, Stuckler D. Comparative performance of private and public healthcare systems in low and middle-income countries: a systematic review. PLoS Med 2012;9(6):E100124. DOI:10.1371/journal.pmed.1001244
    » https://doi.org/10.1371/journal.pmed.1001244
  • 3
    Birkmeyer JD, Dimick JB, Staiger DO. Operative mortality and procedure volume as predictors of subsequent hospital performance. Ann Surg. 2006;243(3):411-7. DOI:10.1097/01.sla.0000201800.45264.51
    » https://doi.org/10.1097/01.sla.0000201800.45264.51
  • 4
    Brand CA, Barker AL, Morello RT, Vitale MR, Evans SM, Scott IA et al. A review of hospital characteristics associated with improved performance. Int J Qual Health Care 2012;24(5):483-94. DOI:10.1093/intqhc/mzs044
    » https://doi.org/10.1093/intqhc/mzs044
  • 5
    Burnett S, Renz A, Wiig S, Fernandes A, Weggelaar AM, Calltorp J et al. Prospects for comparing European hospitals in terms of quality and safety: lessons from a comparative study in five countries. Int J Qual Health Care. 2013;25(1):1-7. DOI:10.1093/intqhc/mzs079
    » https://doi.org/10.1093/intqhc/mzs079
  • 6
    Campbell MJ, Jacques RM, Fotheringham J, Maheswaran R, Nicholl J. Developing a summary hospital mortality index: retrospective analysis in English hospitals over five years. BMJ 2012;344:e1001. DOI:10.1136/bmj.e1001
    » https://doi.org/10.1136/bmj.e1001
  • 7
    Casella GE, Berger RL. Statistical inference. 2nd ed. Pacific Grove: Duxbury; 2002.
  • 8
    Eggleston K, Shen YC, Lau J, Schmid CH, Chan J. Hospital ownership and quality of care: what explains the different results in the literature? Health Econ 2008;17(12):1345-62. DOI:10.1002/hec.1333
    » https://doi.org/10.1002/hec.1333
  • 9
    Fung V, Schmittdiel JA, Fireman B, Meer A, Thomas S, Smider N et al. Meaningful variation in performance: a systematic literature review. Med Care 2010;48(2):140-8. DOI:10.1097/MLR.0b013e3181bd4dc3
    » https://doi.org/10.1097/MLR.0b013e3181bd4dc3
  • 10
    Gala P. A teoria institucional de Douglass North. Rev Econ Polit 2003;23(2):89-105.
  • 11
    Goldstein H, Rasbash J. Improved approximations for multilevel models with binary responses. JR Stat Soc. 1996;59(3):505-13. DOI:10.2307/2983328
    » https://doi.org/10.2307/2983328
  • 12
    Gomes AS, Klück MM, Fachel JMG, Riboldi J. Fatores associados à mortalidade hospitalar na rede SUS do Rio Grande do Sul, em 2005: aplicação de modelo multinível. Rev Bras Epidemiol 2010;13(3):533-42. DOI:10.1590/S1415-790X2010000300016
    » https://doi.org/10.1590/S1415-790X2010000300016
  • 13
    Iezzoni LI. Risk adjustment for performance measurement. In: Smith PC, Mossialos E, Papanicolas I, Leatherman S. Performance measurement for health system improvement: experiences, challenges and prospects. Cambridge: Cambridge University Press; 2009. p. 251-368.
  • 14
    Jarman B, Pieter D, Veen AA, Kool RB, Aylin P, Bottle A et al. The hospital standardised mortality ratio: a powerful tool for Dutch hospitals to assess their quality of care? Qual Saf Health Care 2010;(19):9-13. DOI:10.1136/qshc.2009.032953
    » https://doi.org/10.1136/qshc.2009.032953
  • 15
    Martins M, Blais R, Leite IC. Mortalidade hospitalar e tempo de permanência: comparação entre hospitais públicos e privados na região de Ribeirão Preto, São Paulo, Brasil. Cad Saude Publica 2004;20 Suppl 2:S268-82. DOI:10.1590/S0102-311X2004000800021
    » https://doi.org/10.1590/S0102-311X2004000800021
  • 16
    Martins M. Uso de medidas de comorbidades para predição de risco de óbito em pacientes brasileiros hospitalizados. Rev Saude Publica. 2010;44(3):448-56 . DOI:10.1590/S0034-89102010005000003
    » https://doi.org/10.1590/S0034-89102010005000003
  • 17
    Mohammed MA, Deeks JJ, Girling A, Rudge G, Carmalt M, Stevens AJ et al. Evidence of methodological bias in hospital standardised mortality ratios: retrospective database study of English hospitals. BMJ 2009;338:b780. DOI:10.1136/bmj.b780
    » https://doi.org/10.1136/bmj.b780
  • 18
    Noronha JC. Cobertura universal de saúde: como misturar conceitos, confundir objetivos, abandonar princípios. Cad Saude Publica. 2013;29(5):847-9.
  • 19
    Pitches DW, Mohammed MA, Lilford RJ. What is the empirical evidence that hospitals with higher-risk adjusted mortality rates provide poorer quality care? A systematic review of the literature. BMC Health Serv Res 2007;7(1):91. DOI:10.1186/1472-6963-7-91
    » https://doi.org/10.1186/1472-6963-7-91
  • 20
    Rolim CLRC, Martins M. Qualidade do cuidado ao acidente vascular cerebral isquêmico no SUS. Cad Saude Publica 2011;27(11):2106-16. DOI:10.1590/S0102-311X2011001100004
    » https://doi.org/10.1590/S0102-311X2011001100004
  • 21
    Sanagou M, Wolfe R, Forbes A, Reid CM. Hospital-level associations with 30-day patient mortality after cardiac surgery: a tutorial on the application and interpretation of marginal and multilevel logistic regression. BMC Med Res Methodol 2012;12(1):28. DOI:10.1186/1471-2288-12-28
    » https://doi.org/10.1186/1471-2288-12-28
  • 22
    Santos IS. Evidência sobre o mix público-privado em países com cobertura duplicada: agravamento das iniquidades e da segmentação em sistemas nacionais de saúde. Cienc Saude Coletiva. 2011;16(6):2743-52. DOI:10.1590/S1413-81232011000600013
    » https://doi.org/10.1590/S1413-81232011000600013
  • 23
    Sharabiani MTA, Aylin P, Bottle A. Systematic review of comorbidity indices for administrative data. Med Care. 2012;50(12):1109-18. DOI:10.1097/MLR.0b013e31825f64d0
    » https://doi.org/10.1097/MLR.0b013e31825f64d0
  • 24
    Snijders TAB. Multilevel analysis. In:Lovric M, editor. lnternational encyclopedia of statistical science. Berlin: Springer; 2011. p. 87-82.
  • 25
    Spencer CS, Gaskin DJ, Roberts ET. The quality of care delivered to patients within the same hospital varies by insurance type. Health Aff (Millwoood). 2013;32(10):1731-9. DOI:10.1377/hlthaff.2012.1400
    » https://doi.org/10.1377/hlthaff.2012.1400
  • 26
    Victora CG, Barreto ML, Leal MC, Monteiro CA, Schmidt MI, Paim J et al Health conditions and health-policy innovations in Brazil: the way forward. Lancet. 2011;377(9792):2042-53. DOI:10.1016/S0140-6736(11)60055-X
    » https://doi.org/10.1016/S0140-6736(11)60055-X
  • 27
    Weissman JS, Vogeli C, Levy DE. The quality of hospital care for Medicaid and private pay patients. Med Care 2013;51(5):389-95. DOI:10.1097/MLR.0b013e31827fef95
    » https://doi.org/10.1097/MLR.0b013e31827fef95

  • a
    Machado JP. O arranjo público-privado no Brasil e a qualidade da assistência hospitalar em São Paulo e no Rio Grande do Sul [these]. Rio de Janeiro (RJ): Escola Nacional de Saúde Pública, Fundação Oswaldo Cruz; 2014.
  • b
    Department of Health and Human Services (UEA), Agency for Healthcare Research and Quality. Guide to inpatient quality indicators: quality of care in hospitals: volume, mortality, and utilization. Version 3.1, Rockville, MD: Agency for Healthcare Research and Quality; 2007.

Publication Dates

  • Publication in this collection
    21 July 2016

History

  • Received
    15 Apr 2015
  • Accepted
    1 Sept 2015
Faculdade de Saúde Pública da Universidade de São Paulo São Paulo - SP - Brazil
E-mail: revsp@org.usp.br