J CUa EpidemiolVol. 44, No. 9, pp. 889-894, 1991 Printed in Great Britain. All rights reserved
INEQUALITIES
08954356/91$3.00+ 0.00 Copyright 6 1991Pergamon Press plc
IN HEALTH IN INTENSIVE PATIENTS
CARE
JAIME LATOUR9I* &CENT LOPEZ,* MANUEL RODRIGUEZ,~ ANDREU NOLASCO~and CARLOS ALVAREZ-DARDE~~ Intensive Care Units of ‘Hospital General d’Elx, *Hospital de Sagunt and 3Hospital Lluis Alcanyis de XBtiva and 4Departamento de Salud Comunitaria, Universidad de Alicante, Spain (Received in revised form 26 February 1991)
Abstract-In order to study the possible association between socioeconomic status (SES) and critical care mortality, we examined a cohort of 847 patients over 14 years of age, as they were consecutively admitted to three general intensive care units (ICUs). The patients with low SES (social classes IV and V according to the British Registrar General’s classification) were older (62.0 v 58.5 years old, p < 0.0001) and showed a higher ICU mortality (odds ratio (OR) = 1.61, p = 0.0204) and severity of illness on admission (mean Simplified Acute Physiology Score [SAPS] 9.9 vs 8.7, p = 0.0002) than patients with high SES (social classes I-III). The initial severity of illness differential was detected both in patients admitted from the emergency area and in patients admitted from the general hospitalization ward, suggesting the existence of some kind of preselection procedure related to the SES of the patient. The stepwise logistic regression analysis identified as independent predictive variables of ICU mortality therapeutic effort (measured with the Therapeutic Intervention Scoring System [TISS]), SAPS score, age and hospital, but not SES. The TISS/SAPS ratio according to origin of patients (emergency/general wards) was comparable in the high and low SES. We conclude that there is an inverse relationship between SES and KU mortality. The mortality excess in the low SES patients is largely accounted for by the covariates of the low SES (especially their high age and severity of illness on admission). There is no evidence of a different relative therapeutic effort according to the SES. Critical care factors
Intensive
care
Mortality
INTRODUCTION The existence of social inequalities in health is a subject of continued concern in western developed countries [l-5], despite optimism about their reduction and eventual disappearance [6]. The magnitude of the problem is reflected in the fact that the first target of the World Health Organization’s European regional office for securing Health for All by the year 2000 is the *All correspondence should be addressed to: Dr Jaime Latour Perez, UCI. Hospital General d’Elx, Huertos y Molinos s/n, 03203- Elche, Spain.
reduction
Social class
Socioeconomic
of the inequalities
by at least
25%
[7,81. Possible causes of the social differences in mortality and morbidity include adverse environmental conditions, social alienation, unhealthy lifestyles and inadeqate medical care [g-13]. Unfortunately, studies of the use of hospital services which focus on different socioeconomic status (SES) are rare and controversial. Intensive Care Units (ICUs) seemed to be a suitable place to study this subject. On the one hand, the concentration of a substantial part of hospital mortality in the ICUs 889
JAIMELATOURet
890
made it possible to study the use of hospital resources with mortality as the outcome variable using a reasonable sample size. On the other hand, the availability of highly reliable prognostic indexes for ICU patients allowed us to measure the severity of illness on admission. The aim of the present paper is to answer the following questions: (1) Is there any evidence of a differential ICU mortality according to SES? (2) Are patients from different socioeconomic levels admitted to the ICU with the same degree of severity of illness? (3) Are any differences in ICU mortality attributable to the covariates of the SES and in particular to any differences in initail severity of illness?
al.
[19,20]. The intra-observer concordance was examined by measuring the concordance between the first categorization and the reclassification of the same 30 patients on completion of the study. The association between the exposure of interest and the outcome variable was measured by the odds ratio OR. The null hypothesis of “no association” (OR = 1) was tested using the chi-square test without correction for continuity, or Fisher’s exact test when indicated. In addition, 95% confidence intervals (95%CI) were calculated by Miettinen’s method [21,22]. Differences between the means of continuous variables were examined using the non-paired t-test.
MATERIALS AND METHODS
(a) Patients
The study initially involved 847 patients over 14 years of age who were admitted consecutively to 3 general ICUs (see Appendix). In 23 patients SES could not be ascertained because of the death of the patient soon after admission (14 cases) or administrative errors (9 cases). These patients were excluded from the analysis. The actual study population therefore consisted of 824 patients. (b) Variables The outcome variables chosen were Simplified Acute Physiology Score (SAPS) [14] on admission, a widely used measure of severity of illness based on the physio-pathological impact of the disease, and mortality in the ICU. The “exposure” of interest was low SES (semi-skilled or non-skilled categories of the occupational British Registrar General’s classification)[ 151. In addition, the study controlled the effect of potential confounding factors and/or effect modifiers such as age, sex, SAPS, Therapeutic Intervention Scoring System (TISS) [16-181 (a measure of therapeutic effort), main diagnosis, participating hospital, origin of the patient (emergency area, general ward, surgical area, other), and other social risk factors (occupational status, educational status, income level, marital status).
In order to examine the impact of the patient’s age on the indexes of severity of illness, the TISS and SAPS scores were calculated adjusted for age by covariance analysis. The distortion of the measured association between the risk factor and the outcome variable introduced by each potential confounding factor was controlled using stratified analysis (Mantel-Haenszel chLsquare)[23], according to the principles proposed by Kleinbaum et al. [24]. The effect of the joint confounding was controlled by stepwise logistic regression analysis. The multivariate analysis, the stratified analysis and the analysis of covariance were performed using the BMDP statistical software package [25]. RESULTS
(a) Inter and intra-observer variability
The concordance analysis for evaluation of SES (Table 1) showed that both the intra and the inter-hospital agreement were excellent (Kappa index higher than 0.6). (b) General characteristics of the study population (Tables 2 and 3)
The exposed (low SES) patients were slightly but significantly older than the non-exposed (high SES) patients. The proportion of females was higher in the exposed group. There were no Table 1. Concordance analysis for evaluation of SES
(c) Statistical analysis Agreement between the three principal investigators of the three ICUs on the classification of the SES of the first 30 patients in one of the hospitals was measured by the kappa index
Between-hospitals Intra-hospitals A B C
Kappa index
P
0.71
< 0.0001
0.79 0.85 0.64
< 0.0001 < 0.0001 0.0007
891
Inequalities in Health in ICU Patients Table 2. General characteristics of the study population
Number Age (mean f SD) Gender (% females) TISS (mean k SD) SAPS (mean f SD)
High SES
Low SES
P
418 (51%) 58.5 f 14.8 17.9% 18.9 k 10.5 8.7 + 4.6
406 (49%) 62.0 + 14.9 36.1% 20.1 f 10.6 9.9 + 4.8
< 0.0001 < 0.0001 0.1169 0.0013
major differences according to the source of the patients or their main diagnosis, except for a higher prevalence of patients admitted because of aggravation of chronic obstructive airway disease (p = 0.0057) and a trend for a predominance of toxicologic patients (p = 0.0549) in the exposed group. Both TISS and SAPS were slightly higher in the exposed groups. These differences were significant for SAPS (p = 0.0013) but not for TISS (p = 0.1169). The relative inconsistency of TISS and SAPS could have two extreme explanations: (1) There are no differences in severity of illness on admission, the differences in SAPS are attributable to the higher age (a component of the SAPS index) of the exposed group; (2) The differences in severity of illness are real, the non-significant differences in TISS are due to a low relative therapeutic effort in the low SES group. This dilemma was approached in three different ways: the calculation of age-adjusted TISS and SAPS scores by analysis of covariance, examination of TISS and SAPS scores according to the origin of the patient, and measurement of the TISSjSAPS ratio. Covariance analysis (Table 4) showed that the distortion introduced by the age differences on the severity of illness indexes was of
minor degree. The age-adjusted TISS differences were of borderline statistical significance (p = 0.0665). The stratification according to origin (Table 5) was consistent with other studies [26,27] and showed the higher severity of illness of the patients admitted from the general ward as compared with those admitted from the emergency unit. In the group of patients admitted form the emergency unit, the exposed patients showed higher SAPS and TISS scores (p = 0.0027 and p = 0.0306 respectively). In the group of patients admitted from the general ward, the exposed patients also showed higher values of SAPS (p = 0.0433) and TISS, although the latter was not statistically significant (p = 0.3836). The relative therapeutic effort, measured as the TISS/SAPS ratio, in the exposed and nonexposed patients was comparable. (c) StratiJied analysis (Table 6) The exposed group showed higher ICU mortality than the non-exposed group (crude OR = 1.61, 95%CI between 1.07 and 2.42, p = 0.0204). The stratified analysis showed no evidence of statistical interaction between the SES and any one of the variables considered. However, the association between SES and ICU mortality was stronger in the currently working goup (OR = 3.41, 95%CI of 1.23 and 9.87, p = 0.0151) than in retired group (OR = 1.27, 95%CI of 0.75 and 2.16, p = 0.4132). There was evidence of confounding for several of the variables considered. After adjusting for TISS or SAPS the differences between the exposed
Table 3. Oriein of the oatients and diaenosis High SES
Low SES
Origin: Emergency area General ward Surgical room Other
68.0% 17.3% 12.5% 2.2%
68.5% 21.4% 9.3% 0.1%
Main diagnosis: AMI CAD (different from AMI) Cerebrovascular disease Other cardiovascular COPD Other respiratory disease Gastrointestinal Infectious Accidents/violence Toxic exposure
53.3% 10.0% 1.2% 13.1% 2.1% 1.9% 10.3% 5.3% 4.1% 1.7%
48.3% 9.4% 0.5% 17.0% 6.3% 3.7% 7.1% 4.4% 3.7% 4.2%
p 0.1375
0.1990 0.8676 0.5240 0.1495 0.0057 0.1929 0.1406 0.7179 0.8918 0.0549
Abbreviations-SES: socioeconomic status; AMI: acute myocardial infarction; CAD: coronary artery disease; COPD: chronic obstructive pulmonary disease. CE UP-C
892
JAIME LATOUR et al. Table 4. Analysis
of covariance
Non-exposed
Exposed
p
Mean TISS Crude Adjusted
18.95 18.87
20.14 20.23
0.1169 0.0665
Mean SAPS Crude Adjusted
8.64 8.76
9.90 9.78
0.0013 0.0016
SES could not be ascertained in 23 patients, who were excluded from the analysis. We believe that most of these patients could be classified as low-SES (it is more likely to overload a nonskilled worker than a patient who is a lawyer or a doctor). Given the high mortality of this group (61%), its exclusion from the analysis would probably contribute to underestimating the mortality differential. Secondly, the exposure of interest (SES) can be considered a “soft” variable [28] in which -apart from the concordance analysis-one can expect some degree of misclassification and, consequently, some deviation of the calculated odds ratio towards the null hypothesis. That sort of misclassification bias could explain the possible interaction between SES and occupational status, the association between SES and mortality being stronger in the subgroup “active” (currently working), in which the determination of the SES is more accurate than in the group of “retired” subjects in which the degree of misclassification is probably more important. Our study failed to detect significant differences in mortality once the differences in initial severity of disease were taken into account. The 95% confidence intervals of the adjusted odds ratios (between 0.97 and 2.47 for TISS, and between 0.87 and 2.04 for SAPS) were, however, considerably displaced towards an excess of risk. In addition, the shifted confidence intervals are in line with the logistic analysis which suggested an association between SES and mortality (p = 0.0628) after controlling for TISS and SAPS. Such confidence intervals, which challenge the hypothesis attributing the ICU differences in mortality to the pre-ICU period, can be explained in two ways. First, it is possible that, due to the limited statistical power of the study, the real adjusted differences went undetected (type II error). On the other hand, the possibility of some degree of error in the measurement of the severity of illness (another
and non-exposed group decreased, the adjusted differences appearing to be not significant. This suggests that the excess of mortality in the low-SES group is explained partly by their having more severe illnesses on admission. (d) Stepwise logistic regression analysis
SES was not included as an independent predictor of ICU mortality in the logistic model (Table 7). This suggests that the crude association between SES and ICU mortality is explained by the covariates of SES. DISCUSSION
Our results demonstrate the existence of social inequalities among the patients admitted to the ICU. These inequalities can be seen in the SES-related differential in ICU mortality which, in turn, appears to be explained by the high severity of illness (threshold of access to the ICU) of the low-SES patients. The selection procedure as a result of which the low SES patients are admitted with a mean SAPS score higher than their high-SES counterpart occurs both pre and after the admission, and is not explained by the age differences between the exposed and non-exposed groups. There is no evidence, however, of a differential therapeutic effort in the ICU. The magnitude of the mortality differences detected in our study (OR = 1.61) must be considered with caution. Indeed, there are at least two origins of bias in our estimate. First, Table
5. TISS and SAPS stratified
Origin
General
Operating
room
ward
room
to origin
of patients
Exposed
P
TISS SAPS TISSjSAPS
16.97 f 9.30 8.31 + 4.50 2.44 f 1.72
18.81 * 9.97 9.55 * 4.71 2.25 k 1.52
0.0306 0.0027 0.1733
TISS SAPS TISSjSAPS
20.63 k 10.93 8.79 f 4.73 2.69 f 1.37
22.30 _+ 11.84 10.46 + 5.18 2.68 k 2.53
0.3836 0.0433 0.9704
TISS SAPS TISS/SAPS
25.50 f 9.47 11.84 + 4.05 2.46 k 1.32
24.12 k 7.41 1 I .09 * 4.59 2.49 + 1.24
0.4822 0.6885 0.9249
Index
Emergency
according
Non-exposed
893
Inequalities in Health in ICU Patients Table 6. Stratified analysis Controlled variable
Number of p Test of homogeneity strata
None Hospital Age Sex Marital status Occupational status Income Education TISS SAPS Diagnosis Origin Abbreviations-OR:
1 3 3 2 4 4 2 2 3 3 3 4
0
1 2 3 4
P
0.2803 0.7152 0.3132 0.4690
2.42) 2.62) 2.25) 2.12) 2.23)
0.0204 0.0077 0.0483 0.0713 0.0493
0.1034 0.3748 0.4758 0.4758 0.7133 0.9659 0.4627
1.42 (0.94, 1.36 (0.89, 1.56 (1.04, 1.55 (0.97, 1.33 (0.87, 1.58 (1.06, 1.66 (1.08,
2.16) 2.07) 2.34) 2.47) 2.04) 2.37) 2.53)
0.1084 0.1886 0.0357 0.0722 0.2166 0.0282 0.0208
odds ratio; 95%CI: 95% confidence interval.
Table 7. Multivariate analysis of mortality Step
OR (95%CI) 1.61 (1.07, 1.74 (1.16, 1.51 (1.01, 1.46 (0.98, 1.49 (1.00,
Variables included in the model
p at entry (SES)
Constant + TISS + SAPS + Age + Hospital
0.0246 0.0361 0.0628 0.0887 0.1776
“soft” variable) would cause some degree of “residual confounding” and, thus, a displacement of the adjusted odds ratios away from the null hypothesis. Given the weak association between “severity of illness” and SES, one might expect the bias produced by this residual confounding to be irrelevant [29]. However, a further study with a greater sample size would be necessary to resolve the problem. Unfortunately, our study does not tell us about the exact point at which the preadmission selection procedure takes place. The elucidation of the role played by factors such as the patient’s health behavior, family support, geographical distance from the hospital, quality of the communication between the patient and the medical-nursing staff, etc. requires further studies. Indeed the collaboration between clinicians, epidemiologists, anthropologists and social workers in multidisciplinary research focused on the preselection of patients seems promising.
Acknowledgements-The authors wish to thank Dr. M. Porta and Dr. J. R. Ashton for their revision of the manuscript and their interesting suggestions as well as MS F. Wilson, MS Sarah White and MS K. Spangenberg for their help in translation. This study was partially financed by grant Nos. 87/1527 and 9OjOSlOfrom the Fondo de Investigation Sanitaria de la Seguridad Social (FIS).
REFERENCES 1. Townsend P, Davidson N (Eds). Inequalities in Health. The Black Report. Hardmonsworth, Middlesex: Penguin Books; 1982. 2. Gray AM. Inequalities in health. The Black Report: a summary and comment. Int J Health Serv 1982; 12: 349-380. 3. Morris JN. Social inequalities undiminished. Lancet 1979; i: 87-90. 4. Koskinen S. Time trends in cause-specific mortality by occupational class in England and Wales. International Union for the Scientific Study of Population Conference, Florencia; 1985. 5. Marmot MG, McDowell ME. Mortality decline and widening social inequalities. Lancet 1986; ii: 274. 6. Antonovsky A. Social class, life expectancy and overall mortality. Milbank Mem Fund Q 1967; 45: 31-73. 7. World Health Organization. Targets for Health for All. Copenague: WHO Regional Office for Europe; 1985. 8. Turner J. Europe’s objective. Lancet 1985; ii: 734-735. 9. Kleinman JC, Gold M, Makuc D. Use of ambulatory medical care by the poor: another look at equity. Med Care 1981; 19: 1011-1029. 10. Davis K, Gold M, Makuc D. Access to health care by the poor: does the gap remain? Ann Rev Public Health 1981; 2: 159-182. 11. Newhouse JP, Manning WG, Morris CN, Orr LL, Duan N, Keeler EM, Leibowitz A, Marquis KH, Marquis MS, Phelps CE, Brook RH. Some interim results form a controlled trial of cost sharing in health insurance. N EngI J Med 1981; 305: 1501-1507. 12. Conroy RM, Cahill S, Mulcahy R, Johnson H, Graham IM, Hickey N. The relation of social class to risk factors, rehabilitation, compliance and mortality in survivors of acute coronary heart disease. Stand J Sot Med 1986; 14: 51-56. 13. Cooper RS, Simmons B, Castaner A, Prasad R, Franklin C, Ferliz J. Survival rates and prehospital delay during myocardial infarction among black persons. Am J Cardiol 1986; 57: 208-211. 14. Le Gall JR, Loirat P, Alperovitch A. Simplified acute physiological score of intensive care patients (letter). Lancet 1983; ii: 741. 15. Office of Population Censuses and Surveys. Registrar General’s Decennial Supplement on Occupational Mortality 1979-83. London: HM Stationery Office; 1986. 16. Cullen DJ, Civetta JM, Briggs BA, Ferrara LC. Therapeutic Intervention Scoring System: A method for quantitative comparison of patient care. Crit Care Med 1974; 2: 57-60.
894
JAIMELATOURet al.
17. Keene AR, Cullen DJ. Therapeutic Intervention Scaring System: Update 1983. Crit Care Med 1983; 11: l-3. 18. Cullen DJ, Keene R, Waternaux C, Peterson H. Objective, quantitative measurement of severity of illness in critically ill patients. Crit Care Med 1984; 12: 155-160. 19. Jenicek M, Cleroux R. Epidemiologie Clinique. Clinimetrie. Quebec: Ediseum; 1985. 20. Fleiss SL. Statistical Methods for Rates and hoportions. New York: Wiley; 1981. 21. Meittinen OS. Estimabilitv and estimation in casereferent studies. Am J Epidemiol 1976; 103: 226-235. 22. Fleiss SL. Confidence intervals for the odds ratio in casecontrol studies: state of the art. J Chron Dis 1979; 32: 69-77. 23. Mantel N, Haenszel W. Statistical aspects of the analysis of data from retrospective studies of disease. J Natl Cancer Inst 1959; 22: 719-748. 24. Kleinbaum DG, Kupper LL, Morgenstem H. Epidemiologic Research: Principles and Quantitative Methods. Belmont, CA: Lifetime Learning Publications; 1982. 25. Dixon WJ. BMDP Statistical Software 1981. Berkeley, CA: University of California Press; 1981.
26.
Abizanda R, Jorda R, Valle FX, Chama MA, Cuervo A. Relation entre la procedencia y la mortalidad de 10s pacientes ingresado en una UCI. Med Intensiva 1983; 7: 264-267. 27. Cerda M, Planas M, Pujol U, Roglan A. Valoracion de la procedencia de 10s pacientes atendidos en 10s servicios de medicina intensiva. Med Intensiva 1986; 10: 245-247. 28. Feinstein AR. Clinimetrica. New Haven: Yale University Press; 1987. 29. Kelsey JL, Thompson WD, Evans AS. Methods in Observational Epidemiology. New York: Oxford University Press; 1986.
APPENDIX Participating Hospitals
Hospital General d’Elx (Elche, Spain): Dr J. Latour, Dr J. S. Giner. Hospital de Sagunt (Sagunto, Spain): Dr V. Lopez Camps. Hospital Lluis Alcanyis O(ativa, Spain): Dr M. Rodriguez Serra.
INEQUALITIES
08954356/91$3.00+ 0.00 Copyright 6 1991Pergamon Press plc
IN HEALTH IN INTENSIVE PATIENTS
CARE
JAIME LATOUR9I* &CENT LOPEZ,* MANUEL RODRIGUEZ,~ ANDREU NOLASCO~and CARLOS ALVAREZ-DARDE~~ Intensive Care Units of ‘Hospital General d’Elx, *Hospital de Sagunt and 3Hospital Lluis Alcanyis de XBtiva and 4Departamento de Salud Comunitaria, Universidad de Alicante, Spain (Received in revised form 26 February 1991)
Abstract-In order to study the possible association between socioeconomic status (SES) and critical care mortality, we examined a cohort of 847 patients over 14 years of age, as they were consecutively admitted to three general intensive care units (ICUs). The patients with low SES (social classes IV and V according to the British Registrar General’s classification) were older (62.0 v 58.5 years old, p < 0.0001) and showed a higher ICU mortality (odds ratio (OR) = 1.61, p = 0.0204) and severity of illness on admission (mean Simplified Acute Physiology Score [SAPS] 9.9 vs 8.7, p = 0.0002) than patients with high SES (social classes I-III). The initial severity of illness differential was detected both in patients admitted from the emergency area and in patients admitted from the general hospitalization ward, suggesting the existence of some kind of preselection procedure related to the SES of the patient. The stepwise logistic regression analysis identified as independent predictive variables of ICU mortality therapeutic effort (measured with the Therapeutic Intervention Scoring System [TISS]), SAPS score, age and hospital, but not SES. The TISS/SAPS ratio according to origin of patients (emergency/general wards) was comparable in the high and low SES. We conclude that there is an inverse relationship between SES and KU mortality. The mortality excess in the low SES patients is largely accounted for by the covariates of the low SES (especially their high age and severity of illness on admission). There is no evidence of a different relative therapeutic effort according to the SES. Critical care factors
Intensive
care
Mortality
INTRODUCTION The existence of social inequalities in health is a subject of continued concern in western developed countries [l-5], despite optimism about their reduction and eventual disappearance [6]. The magnitude of the problem is reflected in the fact that the first target of the World Health Organization’s European regional office for securing Health for All by the year 2000 is the *All correspondence should be addressed to: Dr Jaime Latour Perez, UCI. Hospital General d’Elx, Huertos y Molinos s/n, 03203- Elche, Spain.
reduction
Social class
Socioeconomic
of the inequalities
by at least
25%
[7,81. Possible causes of the social differences in mortality and morbidity include adverse environmental conditions, social alienation, unhealthy lifestyles and inadeqate medical care [g-13]. Unfortunately, studies of the use of hospital services which focus on different socioeconomic status (SES) are rare and controversial. Intensive Care Units (ICUs) seemed to be a suitable place to study this subject. On the one hand, the concentration of a substantial part of hospital mortality in the ICUs 889
JAIMELATOURet
890
made it possible to study the use of hospital resources with mortality as the outcome variable using a reasonable sample size. On the other hand, the availability of highly reliable prognostic indexes for ICU patients allowed us to measure the severity of illness on admission. The aim of the present paper is to answer the following questions: (1) Is there any evidence of a differential ICU mortality according to SES? (2) Are patients from different socioeconomic levels admitted to the ICU with the same degree of severity of illness? (3) Are any differences in ICU mortality attributable to the covariates of the SES and in particular to any differences in initail severity of illness?
al.
[19,20]. The intra-observer concordance was examined by measuring the concordance between the first categorization and the reclassification of the same 30 patients on completion of the study. The association between the exposure of interest and the outcome variable was measured by the odds ratio OR. The null hypothesis of “no association” (OR = 1) was tested using the chi-square test without correction for continuity, or Fisher’s exact test when indicated. In addition, 95% confidence intervals (95%CI) were calculated by Miettinen’s method [21,22]. Differences between the means of continuous variables were examined using the non-paired t-test.
MATERIALS AND METHODS
(a) Patients
The study initially involved 847 patients over 14 years of age who were admitted consecutively to 3 general ICUs (see Appendix). In 23 patients SES could not be ascertained because of the death of the patient soon after admission (14 cases) or administrative errors (9 cases). These patients were excluded from the analysis. The actual study population therefore consisted of 824 patients. (b) Variables The outcome variables chosen were Simplified Acute Physiology Score (SAPS) [14] on admission, a widely used measure of severity of illness based on the physio-pathological impact of the disease, and mortality in the ICU. The “exposure” of interest was low SES (semi-skilled or non-skilled categories of the occupational British Registrar General’s classification)[ 151. In addition, the study controlled the effect of potential confounding factors and/or effect modifiers such as age, sex, SAPS, Therapeutic Intervention Scoring System (TISS) [16-181 (a measure of therapeutic effort), main diagnosis, participating hospital, origin of the patient (emergency area, general ward, surgical area, other), and other social risk factors (occupational status, educational status, income level, marital status).
In order to examine the impact of the patient’s age on the indexes of severity of illness, the TISS and SAPS scores were calculated adjusted for age by covariance analysis. The distortion of the measured association between the risk factor and the outcome variable introduced by each potential confounding factor was controlled using stratified analysis (Mantel-Haenszel chLsquare)[23], according to the principles proposed by Kleinbaum et al. [24]. The effect of the joint confounding was controlled by stepwise logistic regression analysis. The multivariate analysis, the stratified analysis and the analysis of covariance were performed using the BMDP statistical software package [25]. RESULTS
(a) Inter and intra-observer variability
The concordance analysis for evaluation of SES (Table 1) showed that both the intra and the inter-hospital agreement were excellent (Kappa index higher than 0.6). (b) General characteristics of the study population (Tables 2 and 3)
The exposed (low SES) patients were slightly but significantly older than the non-exposed (high SES) patients. The proportion of females was higher in the exposed group. There were no Table 1. Concordance analysis for evaluation of SES
(c) Statistical analysis Agreement between the three principal investigators of the three ICUs on the classification of the SES of the first 30 patients in one of the hospitals was measured by the kappa index
Between-hospitals Intra-hospitals A B C
Kappa index
P
0.71
< 0.0001
0.79 0.85 0.64
< 0.0001 < 0.0001 0.0007
891
Inequalities in Health in ICU Patients Table 2. General characteristics of the study population
Number Age (mean f SD) Gender (% females) TISS (mean k SD) SAPS (mean f SD)
High SES
Low SES
P
418 (51%) 58.5 f 14.8 17.9% 18.9 k 10.5 8.7 + 4.6
406 (49%) 62.0 + 14.9 36.1% 20.1 f 10.6 9.9 + 4.8
< 0.0001 < 0.0001 0.1169 0.0013
major differences according to the source of the patients or their main diagnosis, except for a higher prevalence of patients admitted because of aggravation of chronic obstructive airway disease (p = 0.0057) and a trend for a predominance of toxicologic patients (p = 0.0549) in the exposed group. Both TISS and SAPS were slightly higher in the exposed groups. These differences were significant for SAPS (p = 0.0013) but not for TISS (p = 0.1169). The relative inconsistency of TISS and SAPS could have two extreme explanations: (1) There are no differences in severity of illness on admission, the differences in SAPS are attributable to the higher age (a component of the SAPS index) of the exposed group; (2) The differences in severity of illness are real, the non-significant differences in TISS are due to a low relative therapeutic effort in the low SES group. This dilemma was approached in three different ways: the calculation of age-adjusted TISS and SAPS scores by analysis of covariance, examination of TISS and SAPS scores according to the origin of the patient, and measurement of the TISSjSAPS ratio. Covariance analysis (Table 4) showed that the distortion introduced by the age differences on the severity of illness indexes was of
minor degree. The age-adjusted TISS differences were of borderline statistical significance (p = 0.0665). The stratification according to origin (Table 5) was consistent with other studies [26,27] and showed the higher severity of illness of the patients admitted from the general ward as compared with those admitted from the emergency unit. In the group of patients admitted form the emergency unit, the exposed patients showed higher SAPS and TISS scores (p = 0.0027 and p = 0.0306 respectively). In the group of patients admitted from the general ward, the exposed patients also showed higher values of SAPS (p = 0.0433) and TISS, although the latter was not statistically significant (p = 0.3836). The relative therapeutic effort, measured as the TISS/SAPS ratio, in the exposed and nonexposed patients was comparable. (c) StratiJied analysis (Table 6) The exposed group showed higher ICU mortality than the non-exposed group (crude OR = 1.61, 95%CI between 1.07 and 2.42, p = 0.0204). The stratified analysis showed no evidence of statistical interaction between the SES and any one of the variables considered. However, the association between SES and ICU mortality was stronger in the currently working goup (OR = 3.41, 95%CI of 1.23 and 9.87, p = 0.0151) than in retired group (OR = 1.27, 95%CI of 0.75 and 2.16, p = 0.4132). There was evidence of confounding for several of the variables considered. After adjusting for TISS or SAPS the differences between the exposed
Table 3. Oriein of the oatients and diaenosis High SES
Low SES
Origin: Emergency area General ward Surgical room Other
68.0% 17.3% 12.5% 2.2%
68.5% 21.4% 9.3% 0.1%
Main diagnosis: AMI CAD (different from AMI) Cerebrovascular disease Other cardiovascular COPD Other respiratory disease Gastrointestinal Infectious Accidents/violence Toxic exposure
53.3% 10.0% 1.2% 13.1% 2.1% 1.9% 10.3% 5.3% 4.1% 1.7%
48.3% 9.4% 0.5% 17.0% 6.3% 3.7% 7.1% 4.4% 3.7% 4.2%
p 0.1375
0.1990 0.8676 0.5240 0.1495 0.0057 0.1929 0.1406 0.7179 0.8918 0.0549
Abbreviations-SES: socioeconomic status; AMI: acute myocardial infarction; CAD: coronary artery disease; COPD: chronic obstructive pulmonary disease. CE UP-C
892
JAIME LATOUR et al. Table 4. Analysis
of covariance
Non-exposed
Exposed
p
Mean TISS Crude Adjusted
18.95 18.87
20.14 20.23
0.1169 0.0665
Mean SAPS Crude Adjusted
8.64 8.76
9.90 9.78
0.0013 0.0016
SES could not be ascertained in 23 patients, who were excluded from the analysis. We believe that most of these patients could be classified as low-SES (it is more likely to overload a nonskilled worker than a patient who is a lawyer or a doctor). Given the high mortality of this group (61%), its exclusion from the analysis would probably contribute to underestimating the mortality differential. Secondly, the exposure of interest (SES) can be considered a “soft” variable [28] in which -apart from the concordance analysis-one can expect some degree of misclassification and, consequently, some deviation of the calculated odds ratio towards the null hypothesis. That sort of misclassification bias could explain the possible interaction between SES and occupational status, the association between SES and mortality being stronger in the subgroup “active” (currently working), in which the determination of the SES is more accurate than in the group of “retired” subjects in which the degree of misclassification is probably more important. Our study failed to detect significant differences in mortality once the differences in initial severity of disease were taken into account. The 95% confidence intervals of the adjusted odds ratios (between 0.97 and 2.47 for TISS, and between 0.87 and 2.04 for SAPS) were, however, considerably displaced towards an excess of risk. In addition, the shifted confidence intervals are in line with the logistic analysis which suggested an association between SES and mortality (p = 0.0628) after controlling for TISS and SAPS. Such confidence intervals, which challenge the hypothesis attributing the ICU differences in mortality to the pre-ICU period, can be explained in two ways. First, it is possible that, due to the limited statistical power of the study, the real adjusted differences went undetected (type II error). On the other hand, the possibility of some degree of error in the measurement of the severity of illness (another
and non-exposed group decreased, the adjusted differences appearing to be not significant. This suggests that the excess of mortality in the low-SES group is explained partly by their having more severe illnesses on admission. (d) Stepwise logistic regression analysis
SES was not included as an independent predictor of ICU mortality in the logistic model (Table 7). This suggests that the crude association between SES and ICU mortality is explained by the covariates of SES. DISCUSSION
Our results demonstrate the existence of social inequalities among the patients admitted to the ICU. These inequalities can be seen in the SES-related differential in ICU mortality which, in turn, appears to be explained by the high severity of illness (threshold of access to the ICU) of the low-SES patients. The selection procedure as a result of which the low SES patients are admitted with a mean SAPS score higher than their high-SES counterpart occurs both pre and after the admission, and is not explained by the age differences between the exposed and non-exposed groups. There is no evidence, however, of a differential therapeutic effort in the ICU. The magnitude of the mortality differences detected in our study (OR = 1.61) must be considered with caution. Indeed, there are at least two origins of bias in our estimate. First, Table
5. TISS and SAPS stratified
Origin
General
Operating
room
ward
room
to origin
of patients
Exposed
P
TISS SAPS TISSjSAPS
16.97 f 9.30 8.31 + 4.50 2.44 f 1.72
18.81 * 9.97 9.55 * 4.71 2.25 k 1.52
0.0306 0.0027 0.1733
TISS SAPS TISSjSAPS
20.63 k 10.93 8.79 f 4.73 2.69 f 1.37
22.30 _+ 11.84 10.46 + 5.18 2.68 k 2.53
0.3836 0.0433 0.9704
TISS SAPS TISS/SAPS
25.50 f 9.47 11.84 + 4.05 2.46 k 1.32
24.12 k 7.41 1 I .09 * 4.59 2.49 + 1.24
0.4822 0.6885 0.9249
Index
Emergency
according
Non-exposed
893
Inequalities in Health in ICU Patients Table 6. Stratified analysis Controlled variable
Number of p Test of homogeneity strata
None Hospital Age Sex Marital status Occupational status Income Education TISS SAPS Diagnosis Origin Abbreviations-OR:
1 3 3 2 4 4 2 2 3 3 3 4
0
1 2 3 4
P
0.2803 0.7152 0.3132 0.4690
2.42) 2.62) 2.25) 2.12) 2.23)
0.0204 0.0077 0.0483 0.0713 0.0493
0.1034 0.3748 0.4758 0.4758 0.7133 0.9659 0.4627
1.42 (0.94, 1.36 (0.89, 1.56 (1.04, 1.55 (0.97, 1.33 (0.87, 1.58 (1.06, 1.66 (1.08,
2.16) 2.07) 2.34) 2.47) 2.04) 2.37) 2.53)
0.1084 0.1886 0.0357 0.0722 0.2166 0.0282 0.0208
odds ratio; 95%CI: 95% confidence interval.
Table 7. Multivariate analysis of mortality Step
OR (95%CI) 1.61 (1.07, 1.74 (1.16, 1.51 (1.01, 1.46 (0.98, 1.49 (1.00,
Variables included in the model
p at entry (SES)
Constant + TISS + SAPS + Age + Hospital
0.0246 0.0361 0.0628 0.0887 0.1776
“soft” variable) would cause some degree of “residual confounding” and, thus, a displacement of the adjusted odds ratios away from the null hypothesis. Given the weak association between “severity of illness” and SES, one might expect the bias produced by this residual confounding to be irrelevant [29]. However, a further study with a greater sample size would be necessary to resolve the problem. Unfortunately, our study does not tell us about the exact point at which the preadmission selection procedure takes place. The elucidation of the role played by factors such as the patient’s health behavior, family support, geographical distance from the hospital, quality of the communication between the patient and the medical-nursing staff, etc. requires further studies. Indeed the collaboration between clinicians, epidemiologists, anthropologists and social workers in multidisciplinary research focused on the preselection of patients seems promising.
Acknowledgements-The authors wish to thank Dr. M. Porta and Dr. J. R. Ashton for their revision of the manuscript and their interesting suggestions as well as MS F. Wilson, MS Sarah White and MS K. Spangenberg for their help in translation. This study was partially financed by grant Nos. 87/1527 and 9OjOSlOfrom the Fondo de Investigation Sanitaria de la Seguridad Social (FIS).
REFERENCES 1. Townsend P, Davidson N (Eds). Inequalities in Health. The Black Report. Hardmonsworth, Middlesex: Penguin Books; 1982. 2. Gray AM. Inequalities in health. The Black Report: a summary and comment. Int J Health Serv 1982; 12: 349-380. 3. Morris JN. Social inequalities undiminished. Lancet 1979; i: 87-90. 4. Koskinen S. Time trends in cause-specific mortality by occupational class in England and Wales. International Union for the Scientific Study of Population Conference, Florencia; 1985. 5. Marmot MG, McDowell ME. Mortality decline and widening social inequalities. Lancet 1986; ii: 274. 6. Antonovsky A. Social class, life expectancy and overall mortality. Milbank Mem Fund Q 1967; 45: 31-73. 7. World Health Organization. Targets for Health for All. Copenague: WHO Regional Office for Europe; 1985. 8. Turner J. Europe’s objective. Lancet 1985; ii: 734-735. 9. Kleinman JC, Gold M, Makuc D. Use of ambulatory medical care by the poor: another look at equity. Med Care 1981; 19: 1011-1029. 10. Davis K, Gold M, Makuc D. Access to health care by the poor: does the gap remain? Ann Rev Public Health 1981; 2: 159-182. 11. Newhouse JP, Manning WG, Morris CN, Orr LL, Duan N, Keeler EM, Leibowitz A, Marquis KH, Marquis MS, Phelps CE, Brook RH. Some interim results form a controlled trial of cost sharing in health insurance. N EngI J Med 1981; 305: 1501-1507. 12. Conroy RM, Cahill S, Mulcahy R, Johnson H, Graham IM, Hickey N. The relation of social class to risk factors, rehabilitation, compliance and mortality in survivors of acute coronary heart disease. Stand J Sot Med 1986; 14: 51-56. 13. Cooper RS, Simmons B, Castaner A, Prasad R, Franklin C, Ferliz J. Survival rates and prehospital delay during myocardial infarction among black persons. Am J Cardiol 1986; 57: 208-211. 14. Le Gall JR, Loirat P, Alperovitch A. Simplified acute physiological score of intensive care patients (letter). Lancet 1983; ii: 741. 15. Office of Population Censuses and Surveys. Registrar General’s Decennial Supplement on Occupational Mortality 1979-83. London: HM Stationery Office; 1986. 16. Cullen DJ, Civetta JM, Briggs BA, Ferrara LC. Therapeutic Intervention Scoring System: A method for quantitative comparison of patient care. Crit Care Med 1974; 2: 57-60.
894
JAIMELATOURet al.
17. Keene AR, Cullen DJ. Therapeutic Intervention Scaring System: Update 1983. Crit Care Med 1983; 11: l-3. 18. Cullen DJ, Keene R, Waternaux C, Peterson H. Objective, quantitative measurement of severity of illness in critically ill patients. Crit Care Med 1984; 12: 155-160. 19. Jenicek M, Cleroux R. Epidemiologie Clinique. Clinimetrie. Quebec: Ediseum; 1985. 20. Fleiss SL. Statistical Methods for Rates and hoportions. New York: Wiley; 1981. 21. Meittinen OS. Estimabilitv and estimation in casereferent studies. Am J Epidemiol 1976; 103: 226-235. 22. Fleiss SL. Confidence intervals for the odds ratio in casecontrol studies: state of the art. J Chron Dis 1979; 32: 69-77. 23. Mantel N, Haenszel W. Statistical aspects of the analysis of data from retrospective studies of disease. J Natl Cancer Inst 1959; 22: 719-748. 24. Kleinbaum DG, Kupper LL, Morgenstem H. Epidemiologic Research: Principles and Quantitative Methods. Belmont, CA: Lifetime Learning Publications; 1982. 25. Dixon WJ. BMDP Statistical Software 1981. Berkeley, CA: University of California Press; 1981.
26.
Abizanda R, Jorda R, Valle FX, Chama MA, Cuervo A. Relation entre la procedencia y la mortalidad de 10s pacientes ingresado en una UCI. Med Intensiva 1983; 7: 264-267. 27. Cerda M, Planas M, Pujol U, Roglan A. Valoracion de la procedencia de 10s pacientes atendidos en 10s servicios de medicina intensiva. Med Intensiva 1986; 10: 245-247. 28. Feinstein AR. Clinimetrica. New Haven: Yale University Press; 1987. 29. Kelsey JL, Thompson WD, Evans AS. Methods in Observational Epidemiology. New York: Oxford University Press; 1986.
APPENDIX Participating Hospitals
Hospital General d’Elx (Elche, Spain): Dr J. Latour, Dr J. S. Giner. Hospital de Sagunt (Sagunto, Spain): Dr V. Lopez Camps. Hospital Lluis Alcanyis O(ativa, Spain): Dr M. Rodriguez Serra.
