The Importance of Unmeasured Casemix Differences when Evaluating the Outcomes of Premature Infants Delivered at High Volume Neonatal Intensive Care Units SA Lorch, M Baiocchi, C Fager, DS Small Department of Pediatrics Pediatrics, The Children’s Hospital of Philadelphia Division of Statistics Statistics, The Wharton School Background • Neonatal intensive care units (NICUs) developed p in 1970s to manage g premature infants. • Because of the high costs costs, a “regionalized” system was proposed to organize the system. Studies on Regionalization • St Studies di off NICU NICUs suggestt th thatt outcomes improved at high-level NICUs. • Recently, y though, g many y states report p increased de-regionalization of neonatal care in the 1990s. • Few studies of the impact of this change. Limitations in Current Literature • Studies only examine mortality, not other morbidities • No data on state differences in this association • Studies do not account for unmeasured difference in casemix between NICUs. Traditional Casemix Adjustment • Regression R i Analysis A l i • “Adjust away” differences between hospitals Figure 1: The Impact of Regionalization on Neonatal Outcome Hospital Factors (quality) Structure Degree of Regionalization Access to Regional Perinatal Center Neonatal Outcomes in Region - Mortality - Complications - Failure-toRescue - Readmission Patient Factors (Severity, SES) • Examples of unmeasured factors include – Clinical data, data such as fetal heart tracing results – Severity of maternal comorbid conditions Unmeasured Confounding • 2 methods th d – Randomized Trial – Instrumental Variables Approach • Instrument has 2 characteristics: – Influences whether a patient receives the treatment of interest – Does not directly influence the outcome of patient, except by treatment of interest • May require some additional adjustment, for measured factors such as race or SES Instrumental Variables Conceptual Framework Figure 1: The Impact of Regionalization on Neonatal Outcome Hospital Factors Known Pt Factors Structure Degree of Regionalization Access to Regional Perinatal Center Neonatal Outcomes in Region - Mortality - Complications Failure-toto - Failure Rescue - Readmission Instrument A gold standard instrument is a random number used to determine treatment in a clinical trial. Objective T determine To d t i the th association i ti between b t delivery at a high-level NICU and neonatal outcomes before and after adjusting for unmeasured differences in casemix Patient Population • • • • • All deliveries 23-36 23 36 weeks gestational age 1995-2005 Pennsylvania, California, and Missouri “High-level High-level NICU NICU” = – Level III NICU – 50+ deliveries/year of infant with birth weight < 1500 grams Data Sources • Data Sources – Birth and Death Certificates – Maternal and Neonatal Hospital Records • D Data t li linked k d by b name, date d t off birth, bi th race, gender, insurance status, hospital, and zip code • Successful linkage of 98-99% 98 99% of all records. Outcome Measures • Neonatal death • Fetal death • Complications of premature birth – Bronchopulmonary Dysplasia (BPD) – Necrotizing g Enterocolitis (NEC) ( ) – Sepsis – Retinopathy of Prematurity (ROP) – Abdominal Surgery Confounding Variables • Maternal M t l Characteristics Ch t i ti – Age – Chronic medical conditions – Antepartum p complications p – Race – Education • Infant Characteristics – Birth Bi th W Weight i ht and d Gestational G t ti l age – Presence of Congenital Anomalies Instrument • Uses prior data that women tend to deliver at closest hospital. p • IV = difference in distance between high level NICU and nearest delivery hospital for a patient’s zip code • S Smaller ll number: b live li closer l to t high hi h level l l NICU, and should deliver at high level NICU att hi higher h ffrequencies i Data Analysis • Unadjusted U dj t d results lt • Measured casemix control: – Propensity score analysis • Unmeasured casemix control: – Instrumental variables approach, matching patients who live close and far from a high-level NICU on maternal and infant characteristics. • Analyses performed for each state individually. Demographics Pennsylvania California Missouri HL NICU Other NICU HL NICU Other NICU HL NICU Other NICU Differential Travel Ti Time (min) ( i ) 6 97 6.97 21 81 21.81 3 45 3.45 14 52 14.52 14 99 14.99 39 50 39.50 Birth Weight 2,473 2,724 2,716 2,935 2,649 2,853 Gestational Age 34.66 35.72 34.85 35.42 34.71 35.32 White 0.64 0.78 0.59 0.60 0.78 0.75 Black 0 22 0.22 0 09 0.09 0 10 0.10 0 06 0.06 0 19 0.19 0 23 0.23 Asian 0.01 0.01 0.09 0.09 0.02 0.02 Other 0.03 0.03 0.20 0.23 0.01 0.01 0.81 0.87 0.88 0.92 0.86 0.91 Preterm Labor 0 49 0.49 0 40 0.40 0 34 0.34 0 22 0.22 0 37 0.37 0 27 0.27 PROM 0.21 0.15 0.12 0.09 0.16 0.11 PIH 0.12 0.08 0.08 0.05 0.10 0.08 Race Singleton Birth Maternal Complication Strength g of Instrument Impact of Unmeasured Casemix on Outcomes, O t PA 1995 1995-2005 2005 Results, Unadjusted RD Neonatal Death Fetal Death 1 09% 1.09% 0.76% Results, adjusted for measured factors Results, adjusted for measured and unmeasured factors RR RD RR RD RR 1 88 1.88 -0.10% ( 0 21% 0 (-0.21%, 0.02%) 02%) 0.90 (0 78 1 (0.78, 1.02) 02) -0.99% ( 1 36% -0.61%) (-1.36%, 0 61%) 0.50 (0 32 0 (0.32, 0.69) 69) 2.73 -0.02% (-0.1%, 0.06%) 0.94 (0.71, 1.17) -0.05% (-0.29%, 0.19%) 0.92 (0.55, 1.29) Positive Risk Difference (RD) = higher mortality at highlevel NICU Negative Risk Difference = lower mortality at high-level NICU Impact of Unmeasured Casemix on Outcomes, O t PA 1995 1995-2005 2005 Results, Unadjusted RD BPD NEC Fungal Sepsis Bacterial Sepsis ROP ROP g y Surgery Laparotomy RR Results, adjusted for measured factors Results, adjusted for measured and unmeasured factors RD RR RD RR 1.86% 3.83 0.49% (0.37%, 0.6%) 1.69 (1.52, 1.85) 0.05% (-0.31%, 0.41%) 1.03 (0.83, 1.22) 0.63% 2.98 0.09% (0%, 0.17%) 1.25 (1.00, 1.49) 0.16% (-0.09%, 0.41%) 1.25 (0.86, 1.65) 0.77% 2.97 0.43% (0.32%, 0.54%) 2.07 (1.80, 2.33) 0.50% (0.24%, 0.77%) 1.77 (1.36, 2.17) 2.48% 1.94 0.19% (-0.03%, 0.41%) 1.07 (0.99, 1.15) 0.96% (0.40%, 1.51%) 1.28 (1.12, 1.44) 0.77% 3.28 0 5% 0.15% (0.06%, 0.24%) 1.44 (1.19, 1.70) -0.10% 0 0% (-0.35%, 0.16%) 08 0.87 (0.54, 1.21) 0.12% 3.79 0.03% ((-0.01%, 0.06%)) 1.58 ((0.87, 2.29)) -0.10% ((-0.21%, 0.02%)) 0.38 ((-0.33, 1.10)) 0.18% 1.97 0.02% (-0.03%, 0.07%) 1.18 (0.77, 1.59) -0.05% (-0.21%, 0.11%) 0.82 (0.26, 1.38) Unadjusted and Adjusted Outcomes Conclusions • Unmeasured differences in casemix contribute to lack of difference in outcomes between hospitals. • Different results depending on the outcome studied. • Large differences between states. Implications for Policy • Ignoring unmeasured casemix differences may provide support for a deregionalized system of perinatal care. • Delivery y location remains important to optimize perinatal outcomes. • The instrumental variables approach may improve the assessment of outcomes while awaiting future EHR methods to control for differences in casemix. Thanks • Departments of Health – California – Pennsylvania – Missouri • AHRQ R01 HS015696