Variation in pediatric intensive care therapies and outcomes by race, gender, and insurance status*

Size: px
Start display at page:

Download "Variation in pediatric intensive care therapies and outcomes by race, gender, and insurance status*"


1 Feature Articles Variation in pediatric intensive care therapies and outcomes by race, gender, and insurance status* Adriana M. Lopez, MD; John M. Tilford, PhD; K. J. S. Anand, MBBS, Dphil; Chan-Hee Jo, PhD; Jerril W. Green, MD; Mary E. Aitken, MD, MPH; Debra H. Fiser, MD Context: The differential allocation of medical resources to adult patients according to characteristics such as race, gender, and insurance status raises the serious concern that such issues apply to critically ill children as well. Objective: This study examined whether medical resources and outcomes for children admitted to pediatric intensive care units differed according to race, gender, or insurance status. Design: An observational analysis was conducted with use of prospectively collected data from a multicenter cohort. Data were collected on 5,749 consecutive admissions for children from three pediatric intensive care units located in large urban children s hospitals. Participants: Children aged <18 years admitted over an 18- month period beginning in June 1996 formed the study sample. Main Outcome Measures: Hospital mortality, length of hospital stay, and overall resource use were examined in relation to severity of illness. Standardized ratios were formed with generalized regression analyses that included the Pediatric Index of Mortality for risk adjustment. Results: After adjustment for differences in illness severity, standardized mortality ratios and overall resource use were similar with regard to race, gender, and insurance status, but uninsured children had significantly shorter lengths of stay in the pediatric intensive care unit. Uninsured children also had significantly greater physiologic derangement on admission (mortality probability, 8.1%; 95% confidence interval [CI], ) than did publicly insured (3.6%; 95% CI, ) and commercially insured patients (3.7%; 95% CI, ). Consistent with greater physiologic derangement, hospital mortality was higher among uninsured children than insured children. Conclusions: Risk-adjusted mortality and resource use for critically ill children did not differ according to race, gender, or insurance status. Policies to expand health insurance to children appear more likely to affect physiologic derangement on admission rather than technical quality of care in the pediatric intensive care unit setting. (Pediatr Crit Care Med 2006; 7:2 6) KEY WORDS: treatment disparities; critically ill children; mortality; outcomes; risk adjustment Racial and ethnic disparities in the provision of health care pose a special problem because these differences often are viewed as a product of discrimination. Indeed, the Institute of Medicine reported that racial and ethnic minorities receive a lower quality of health care than nonminorities, even after controlling for confounding factors such as socioeconomic status and insurance status (1). Concerns over a lower quality of health care for minorities because of racial and ethnic disparities led to the recent introduction *See also p. 86. From the Department of Pediatric Critical Care, University of Texas Health Science Center at San Antonio (AML), San Antonio, TX; and Department of Pediatrics, College of Medicine, University of Arkansas for Medical Sciences and Arkansas Children s Hospital (JMT, KJSA, C-HJ, JWG, MEA, DHF), Little Rock, AR. Supported in part by grant HS09055 from the Agency for Healthcare Research and Quality, in collaboration with the Health Resources and Services Administration, Maternal and Child Health Bureau. The authors do not have a financial interest in publication of federal legislation. This legislation is aimed at reducing disparities affecting minority and underserved populations, especially in the areas of cancer, asthma, and human immunodeficiency virus/ acquired immunodeficiency syndrome (2). Although less studied, racial disparities have been documented in the allocation of resources to critically ill patients. After controlling for hospital effects and insurance status, Yergan et al. (3) found that blacks were admitted to intensive care units less often than whites with the of the article. Address requests for reprints to: John M. Tilford, PhD, Associate Professor, Center for Applied Research and Evaluation, Arkansas Children s Hospital, 800 Marshall Street, Little Rock, AR Copyright 2006 by the Society of Critical Care Medicine and the World Federation of Pediatric Intensive and Critical Care Societies DOI: /01.PCC same diagnosis. In addition, black patients who were admitted to intensive care units were shown to have shorter lengths of stay and less resource use (4). Disparities in health care are not limited to race and ethnicity; a number of studies have documented disparities according to insurance status and gender. For example, one study found reduced lengths of stay, hospital charges, and rates of mechanical ventilation according to insurance status (5). Other studies have shown uninsured patients to be at a greater risk for adverse events due to negligence and revealed lower resource utilization and higher mortality in this population (6, 7). Studies also have demonstrated that men were much more likely than women to undergo invasive procedures in the intensive care setting, despite women having a higher measured severity of illness (8, 9). Finally, women with coronary artery disease were more likely to be discharged with a missed diagnosis of acute myocardial infarction, and once admitted, they were referred for 2 Pediatr Crit Care Med 2006 Vol. 7, No. 1

2 coronary bypass surgery later in the treatment course than were men (10, 11). Almost all of the studies documenting disparities in health care treatments, however, were focused on adult populations. Although disparities in treatment according to patient characteristics are an important policy concern, it is especially important to know whether disparities exist in pediatric populations. One pediatric population of special interest is critically ill children requiring intensive care services, since these children are at an increased risk of death. To address whether race, gender, or insurance status significantly alters the allocation of therapies to critically ill children, this study examined resource use and outcomes in a large cohort of children admitted to the pediatric intensive care units (PICUs) of three children s hospitals. METHODS Data Collection and Main Outcome Measures. Data for this study were obtained from three PICUs. The PICUs were not selected randomly; the intent was to make the sample more representative by using multiple sites. All three PICUs were located in children s hospitals, were associated with fellowship and residency training programs, and were directed by pediatric intensivists. Despite these similarities, the PICUs differed by important organizational characteristics such as trauma center designation status and 24-hr in-house attending coverage. Consecutive admissions to each of the three units were recorded over an 18-month period, from June 1996 to December 1997 (n 5,749). Patients with multiple admissions were included and regarded as separate admissions. Each unit had a designated nurse who abstracted medical records and entered data into a programmed database supported with range and logic checks to reduce data entry errors. Approximately 7% of records at each institution were re-entered as part of the data abstraction and data management quality control system. Data were collected on the main outcome measures (hospital mortality, length of stay [LOS], and resource use) as well as clinical information to measure the severity of illness, patient demographics, and insurance status. status was coded as private, public (Medicaid and other public forms of insurance), or self-pay (uninsured). status was recorded on admission to the PICU to distinguish patients who might qualify for Medicaid or other public insurance because of the admission. Clinical measures were identified upon hospital arrival or within 1 hr following admission to the PICU. Clinical measures and diagnostic information were used to form the Paediatric Index of Mortality (PIM2) score (12). The PIM2 score is calculated on the basis of ten variables including indicators of highrisk and low-risk diagnosis. Data to calculate the PIM2 should be obtained within 1 hr of admission to the PICU on consecutive admissions. Each of the ten variables can be entered into a logit formula to form the PIM2 score, and that score can be converted to a probability of mortality by means of standard methods based on logistic regression analysis (13). We report the probability of mortality from the PIM2 score and use this probability in all analyses. Resource use measures on all admissions were collected daily and were based on the Therapeutic Intervention Scoring System (TISS) (14, 15) adapted for pediatrics (16). The TISS contains 72 therapies routinely used in ICUs that are weighted according to their impact on resource utilization. TISS items were recorded for every patient daily to permit an analysis of the individual therapies received, the total TISS score for the entire admission, and the average TISS score per admission day. Outcome was based on whether the patient died in the PICU. Although some patients are placed on comfort care and transferred to other institutions or hospital wards, the vast majority of pediatric patients who die remain in the PICU. The Human Research Advisory Committee at the University of Arkansas for Medical Sciences approved the study protocol. The requirement for informed consent was waived by exemption. Statistical Analysis. Standardized ratios for hospital mortality, LOS, and resource use were calculated by dividing observed with expected outcomes and then were assessed by race, gender, and insurance status. Expected outcomes were generated by stepwise generalized linear regression models where the PIM2 score was customized to the current database. Methods for customizing the PIM2 score followed recent work by Glance et al (17). A number of studies in the literature describe the rationale for customizing severity measures (17 20). The basic premise for customization is the need to generate observedto-expected values that are identical at the mean over the entire database. Failure to customize severity measures can result in observed-to-expected ratios below 1 for all race, gender, or insurance subsets (21). The PIM2 score was included in the mortality regression model with use of a quadratic specification to improve model fit. Other studies have used quadratic specifications for similar reasons (17, 22). Additional variables to achieve adequate model fit also were included. A description of these variables has been reported previously (23). Logistic regression analysis to predict hospital mortality involved the use of both calibration and discrimination statistics (24, 25). Separate models were used to predict LOS and average TISS scores with use of negative binomial (26, 27) and linear Table 1. Characteristics of the patients Characteristic n (%) Male 3,305 (57.5) Female 2,444 (42.5) Age, yrs Neonate 254 (4.4) 1 1,354 (23.6) 1 5 1,683 (29.3) ,759 (30.6) (12.2) Asian 237 (4.1) Black 1,925 (33.5) Caucasian 3,112 (54.1) Hispanic 372 (6.5) Other 103 (1.8) Status Private 2,549 (44.3) Public 2,397 (41.7) None (self-pay) 789 (13.7) Survival Survived 5,536 (96.3) Died 213 (3.7) PICU A 1,993 (34.7) B 2,202 (38.3) C 1,554 (27.0) Total sample 5,749 (100.0) PICU, pediatric intensive care unit. regression analysis. Average TISS scores were used to measure resource use to avoid statistical issues associated with the skewed distribution inherent in total TISS scores. All analyses used methods to adjust for the clustering of patients at the PICU level (28). The standardized ratios generated by dividing observed with expected outcomes form the basis for assessing disparities in resource use and outcomes according to patients characteristics. Confidence intervals (CIs) for the standardized mortality ratios were calculated with methods for cohort studies (29), whereas bootstrap methods were used for the resource use and length-of-stay estimates. A 95% CI strictly less than 1.0 indicates a better than expected outcome, whereas a CI strictly above 1.0 indicates a worse than expected outcome. CIs that include the value 1.0 indicate no difference in outcome. RESULTS Over the study period, data on 5,749 admissions to the three PICUs were abstracted. Table 1 provides characteristics of the children admitted to the three PICUs. Examination of table 1 indicates that the majority of patients were male (57.5%) and 87.6% were of either black or Caucasian race. The percentage of children admitted with private insurance (44.3%) was similar to the percentage with public insurance (41.7%). The per- Pediatr Crit Care Med 2006 Vol. 7, No. 1 3

3 Table 2. Distribution of length of stay and resource use measures Measure Mean (SD) Median Range Length of stay, days 4.2 (5.8) Total TISS score 70.8 (142.5) Average daily TISS score 13.1 (8.0) TISS, Therapeutic Intervention Scoring System. Table 3. Mortality rates and severity scores by race, gender, and insurance status Characteristic Mortality Rate (95% CI) PIM2 Probability (95% CI) Asian 5.1 ( ) 5.2 ( ) Black 3.2 ( ) 3.8 ( ) Caucasian 3.7 ( ) 4.1 ( ) Hispanic 5.6 ( ) 5.9 ( ) Other 3.9 ( ) 5.7 ( ) Female 3.4 ( ) 3.8 ( ) Male 3.9 ( ) 4.5 ( ) status Private 2.7 ( ) 3.7 ( ) Public 3.3 ( ) 3.6 ( ) None (self-pay) 8.1 ( ) 7.3 ( ) Total sample 3.7 ( ) 4.2 ( ) CI, confidence interval; PIM2, Pediatric Index of Mortality. centage of children with self-pay status (uninsured) was similar to the national average of 13.8% in 1995 (30). Table 2 provides data on LOS and resource use outcomes for the overall sample. This table indicates that average LOS and total resource use have standard deviations in excess of their means because of the skewed distribution associated with these measures. Thus, multivariate analysis of LOS used negative binomial models, and resource use was assessed by dividing total TISS scores by LOS. The resulting distribution tended toward normality and was assessed with linear regression models. Table 3 provides data on mortality and severity of illness at admission to the PICU with regard to PIM2 probabilities of mortality. Over the entire sample, the PIM2 score predicted a mortality rate of 4.2%, vs. the actual rate of 3.7%. The predicted mortality rate also exceeded the actual mortality rate for all racial groupings and by males and females. The PIM2 probabilities indicated a much higher severity of illness on admission for the selfpay group. Probability of mortality for the self-pay patients was twice that for insured patients; PIM2 mortality probability was 8.1% (95% CI, ) for selfpay patients, vs. 3.6% (95% CI, ) for the publicly insured and 3.7% (95% CI, ) for the commercially insured. Actual mortality among the selfpay patients also was higher than among insured patients. Table 3 illustrates the need to customize scoring systems to current databases to make inferences about technical quality of care by race, gender, or insurance status. Customizing the PIM2 score with other covariates to the current database proved successful. Calibration was excellent, with a Hosmer- Lemeshow chi-square of 9.06 and a corresponding p value of The area under the receiver-operator characteristic curve for the model was excellent as well, at Figure 1 shows the use of mechanical ventilation according to insurance status. Rates of mechanical ventilation illustrate a step-function, with privately insured children having the lowest rate of ventilation, followed by publicly insured children and uninsured children. The stepfunction for use of mechanical ventilation is the opposite of the wellknown relationship between insurance status and use of primary care (31). Tables 4 6 provide standardized ratios for mortality, LOS, and resource use by race, gender, and insurance status as estimated from observed to expected outcomes, where expected outcomes were estimated by stepwise regression analyses. There were no differences in the standardized mortality ratios, by race, gender, or insurance status, because all of the 95% CIs contained the value 1.0. In Figure 1. Ventilator support and well-child visits in relation to insurance status. Data source: author s calculations and Reference 31. Table 4. Standardized mortality ratios by race, gender, and insurance status Characteristic SMR 95% CI Black White Other Private Public None Male Female Total sample SMR, standardized mortality ratio; CI, confidence interval. Table 5. Standardized length of stay ratios by race, gender, and insurance status Characteristic Standardized LOS Ratio 95% CI Black White Other Private Public None Male Female Total sample LOS, length of stay; CI, confidence interval. contrast to the standardized mortality ratios, the standardized LOS ratios exhibited differences by insurance status: children without insurance on admission to the PICU had a significantly shorter LOS (0.86; 95% CI, ), whereas the publicly insured had an increased LOS (1.08; 95% CI, ). Finally, the standardized resource use ratios indicated minor variations by race, with a 4% difference in resource use between black 4 Pediatr Crit Care Med 2006 Vol. 7, No. 1

4 Table 6. Standardized resource use ratios by race, gender, and insurance status Characteristic children and white children, consistent with findings in previous studies of adults (4). However, both CIs were just outside the range of standard levels of significance. DISCUSSION Standardized Resource Use Ratio 95% CI Black White Other Private Public None Male Female Total sample CI, confidence interval. Brook et al. (32) once asked what would be the effect on society as the currently disadvantaged populations learn through the public release of information that they are receiving inferior care. This same concern prompted us to undertake the present study. If a parent with knowledge of health care disparities brings a child into the hospital, there could be a sense of hesitancy, mistrust, or concern that the child might receive inferior care on the basis of race, gender, socioeconomic status, or insurance status. Will poor outcomes be perceived as a consequence of discrimination? This study examined variations in PICU therapies and outcomes by race, gender, and insurance status. Few differences were associated with race or gender. Examination of variations by insurance status revealed that far more serious illness and injuries were associated with uninsured children admitted to the PICU. Mortality rates and PIM2 scores for uninsured children were much higher than for privately or publicly insured children (Table 2). Further investigation of this finding produced little evidence indicating a lower quality of care for uninsured children relative to privately insured or publicly insured children. After adjustment for physiologic status on admission to the PICU and hospital effects, the standardized mortality ratios showed no significant differences. Concerns about social discrimination as a cause of poor outcomes in PICUs are not supported by these data. However, some public concern may be raised over the finding that uninsured children have significantly shorter LOS. Two explanations may account for this finding. Uninsured children present to the PICU with vastly different diagnoses and die at a higher rate. Both differential diagnoses and death during the early PICU course may account for a shorter LOS for uninsured children. Our adjusted LOS models accounted for differential diagnoses and mortality following similar work in adult populations (5). We further tested whether standardized LOS ratios for uninsured children still differed if based only on children who survived. Results were similar, whether all uninsured children or only uninsured children who survived were included (relative risk, 0.88; 95% CI, ). The study has several limitations. Data were collected from a convenience sample of three PICUs that were all at teaching children s hospitals. Our findings may not be applicable to patients who do not receive care in large teaching hospitals or in freestanding children s hospitals. Previous studies have documented that PICUs with greater volumes and fellowship programs have lower PICU mortality risk (33,34). Our findings, however, are similar to recent findings from population-based data indicating mortality risk was elevated in critically ill, uninsured children following traumatic brain injury (35). Finally, the Hispanic and Asian populations were underrepresented. Reported disparities in the use of cardiovascular procedures are not limited to blacks but have been demonstrated in Asians and Hispanics as well (36). Our findings that uninsured children present with vastly different physiologic derangements should generate some interest in investigating the etiology of these differences. We demonstrated that need for mechanical ventilation is related to insurance status, with privately insured children having the lowest rate of mechanical ventilation and uninsured children having the highest rate. The findings may be related to well-known relationships between the use of primary care (31) and the quality of primary care (37). Federal and state initiatives have expanded health insurance for children of lower socioeconomic status, with the goal of increasing their access to care and ultimately improving their health status (38). This study illustrates the value of using PICU outcomes as an indicator of health status for different systems of care. Understanding the consequences of being uninsured (39) requires information on whether PICU outcomes are influenced by technical quality of care or by the degree of physiologic derangement on admission. The findings suggest health insurance expansions may have greater impact on the severity of physiologic derangement on admission than on the technical quality of care. Public discourse should reflect the importance of research on disparities with regard to critically ill children to prevent unwarranted parental reactions associated with poor outcomes. REFERENCES 1. Unequal Treatment: Confronting Racial and Ethnic Disparities in Health Care. Washington, DC, Institute of Medicine, Frist WH: Overcoming disparities in U.S. health care. Health Aff 2005; 24: Yergan J, Flood AN, Logerfo J, et al: Relationship between patient race and the intensity of hospital services. Med Care 1987; 25: Williams JF, Zimmerman JE, Wagner DP, et al: African-American and white patients admitted to the intensive care unit: Is there a difference in therapy and outcome? Crit Care Med 1995; 23: Rapoport J, Gelbach S, Lemeshow S, et al: Resource utilization among intensive care patients: Managed care vs traditional insurance. Arch Intern Med 1992; 152: Burstin HR, Lipsitz SR, Brennan TA: Socioeconomic status and risk for substandard medical care. JAMA 1992; 268: Haas JS, Goldman L: Acutely injured patients with trauma in Massachusetts: Differences in care and mortality, by insurance status. Am J Public Health 1994; 84: Giles WH, Anda RF, Casper ML, et al: and sex differences in rates of invasive cardiac procedures in US hospitals: Data from the National Hospital Discharge Survey. Arch Intern Med 1995; 155: Valentin A, Jordan B, Lang T, et al: related differences in intensive care: A multiple-center cohort study of therapeutic interventions and outcome in critically ill patients. Crit Care Med 2003; 31: Khan SS, Nessim S, Gray R, et al: Increased mortality of women in coronary artery bypass surgery: Evidence for referral bias. Ann Intern Med 1990; 112: Pope JH, Aufderheide TP, Ruthazer R, et al: Missed diagnoses of acute cardiac ischemia in the emergency department. N Engl J Med 2000; 342: Slater A, Shann F, Pearson G: PIM2: A re- Pediatr Crit Care Med 2006 Vol. 7, No. 1 5

5 vised version of the Paediatric Index of Mortality. Intensive Care Med 2003; 29: Hosmer DW, Lemeshow S. Applied Logistic Regression. New York, John Wiley & Sons, Malstam J, Lind L: Therapeutic intervention scoring system (TISS): A method for measuring workload and calculating cost in the ICU. Acta Anaesthesiol Scand 1992; 36: Cullen DJ, Nemeskal AR. Therapeutic intervention scoring system (TISS). In: Farmer JC (Ed). Problems in Critical Care: Prognostic Scoring Systems in the ICU. Philadelphia, JB Lippincott, 1989, pp Yeh TS, Pollack MM, Holbrook PR, et al: Assessment of pediatric intensive careapplication of the therapeutic intervention scoring system. Crit Care Med 1982; 10: Glance LG, Osler TM, Dick A: Identifying quality outliers in a large, multiple-institution database by using customized versions of the Simplified Acute Physiology Score II and the Mortality Probability Model. Crit Care Med 2002; 30: Zhu BP, Lemeshow S, Hosmer DW, et al: Factors affecting the performance of the models in the Mortality Probability Model II system and strategies of customization: A simulation study. Crit Care Med 1996; 24: Le Gall JR, Lemeshow S, Leleu G, et al: Customized probability models for early severe sepsis in adult intensive care patients. JAMA 1995; 273: Teres D, Rapoport J, Lemeshow S, et al: Effects of severity of illness on resource use by survivors and nonsurvivors of severe sepsis at intensive care unit admission. Crit Care Med 2002; 30: Tilford JM, Parker JG: A gender bias in the allocation of intensive care unit resources? Crit Care Med 2003; 31: Pollack MM, Cuerdon TT, Patel KM, et al: Impact of quality-of-care factors on pediatric intensive care unit mortality. JAMA 1994; 272: Tilford JM, Simpson PM, Yeh T, et al: Variation in therapy and outcome for pediatric head trauma patients. Crit Care Med 2001; 29: Hadorn DC, Keeler EB, Rogers WH, et al: Assessing the Performance of Mortality Prediction Models. Santa Monica, CA, RAND, Lemeshow S, LeGall JR: Modeling the severity of illness of ICU patients. JAMA 1994; 272: Brown KL, Ridout DA, Goldman AP, et al: Risk factors for long intensive care unit stay after cardiopulmonary bypass in children. Crit Care Med 2003; 31: Lage MJ, Barber BL, McCollam PL, et al: Impact of abciximab versus eptifibatide on length of hospital stay for PCI patients. Catheter Cardiovasc Interv 2001; 53: Mancl LA, Leroux BG: Efficiency of regression estimates for clustered data. Biometrics 1996; 52: Sasieni P: Tabulation of observed/expected ratios and confidence intervals: Update. Stata Technical Bulletin 1996;5: Health Coverage Status of All Persons by Selected Characteristics: Washington, DC, U.S. Census Bureau, Short PF, Lefkowitz DC: Encouraging preventive services for low-income children: The effect of expanding Medicaid. Med Care 1992; 30: Brook RH, Kamberg CJ, McGlynn EA: Health system reform and quality. JAMA 1996; 276: Tilford JM, Simpson PM, Lensing S, et al: Volume-outcome relationships in pediatric intensive care units. Pediatrics 2000; 106: Pollack MM, Patel KM, Ruttiman UE: Pediatric critical care training programs have a positive effect on pediatric intensive care mortality. Crit Care Med 1997; 25: Tilford JM, Aitken ME, Anand KJS, et al: Hospitalizations for critically ill children with traumatic brain injuries: A longitudinal analysis. Crit Care Med 2005; 33: Carlisle DM, Leake BD, Shapiro MF: Racial and ethnic differences in the use of invasive cardiac procedures among cardiac patients in Los Angeles County, 1986 through Am J Public Health 1995; 85: Thompson JW, Ryan KW, Pinidiya SD, et al: Quality of care for children in commercial and Medicaid managed care. JAMA 2003; 290: Hughes DC, Ng S: Reducing health disparities among children. Future Child 2003; 13: Hadley J: Sicker and poorer: The consequences of being uninsured: A review of the research on the relationship between health insurance, medical care use, health, work, and income. Med Care Res Rev 2003; 60(2 Suppl):3S 75S. 6 Pediatr Crit Care Med 2006 Vol. 7, No. 1