analyses of clinical data can be more easily facilitated in single centers as well as in emerging collaborative arrangements.61 Detailed clinical data are available, although accurate abstraction depends on consistency in documenting the start and resolution of illnesses, and limiting diagnostic variability between centers for common illnesses such as apnea.62,63 In addition to difficulties inherent in obtaining research data from a clinical chart, challenges to study design include 838 Lagatta et al identification of interventions and outcomes, a study population and follow-up interval, and a plan for active versus passive data capture.64 One of the most established sources of secondary analysis using medical records data in neonatology comes from the Pediatrix Medical Group, whose Clinical Data Warehouse automatically facilitates export of de-identified, discrete data elements from patient charts.65 Several comparative effectiveness studies have come from this group, including comparisons of adverse events after differing preparations of surfactant32 and different empiric antibiotic regimens.66 In many fields of medicine, administrative data taken from billing claims are commonly used for CER. Compared with patient registry or electronic health record data, these data sources are more likely to be limited by lack of completeness of the listed diagnoses. Acute and particularly surgical conditions are more likely to be coded appropriately. Neonatology, as in many other fields, has some discrepancies between clinical diagnoses and coding terminology for common diseases, such as bronchopulmonary dysplasia and respiratory distress syndrome, which limits the clinical detail needed to design CER studies.67 Public use data files, such the National Inpatient Sample and its associated Kids Inpatient Database available from the Agency for Healthcare Research and Quality, have similar limitations in diagnosis availability because they are derived from billing data. However, billing files are a potential source of post-NICU follow-up data on rehospitalizations or costs of outpatient care, which could provide useful outcome measures. Linked data sets that pair the longitudinal data collection of billing data with the appropriate amount of neonatal coding accuracy, such as the California Perinatal Quality Care Collaborative,68 the Kaiser Permanente Neonatal Minimum Data Set,69 or the Children’s Hospitals Association Neonatal Database,63 could be used to conduct this type of research. CHALLENGES IN NEONATAL COMPARATIVE EFFECTIVENESS RESEARCH Comparing Treatment with Placebo: Efficacy Versus Effectiveness Perhaps more commonly than in other clinical fields of medicine, for many diseases of prematurity, a potential “standard of care” is observation without intervention. There are no noted differences in mortality or morbidity whether a hypotensive extremely low birth weight infant with reasonable end-organ perfusion receives a vasopressor or clinical observation alone.59 Thus, in some circumstances, comparing an intervention with observation alone could be considered an effectiveness study. Historically, some treatments that were adopted as standard practice without controlled trials, such as bicarbonate for metabolic acidosis, turned out to be worse than placebo.70 Although there are other instances in medicine where observation alone is a viable treatment option (early stage prostate cancer), very few other specialties see an individual patient grow 5- to 7-fold over the course of a single hospitalization, making observation without intervention potentially a more relevant therapeutic option in neonatology than in other fields. Defining and Measuring Patient-Centered Outcomes The Institute of Medicine has identified “patient-centered outcomes” as outcomes that are directly relevant to stakeholders, rather than proxy measures. Depending on the research question, a stakeholder could be a patient or parent, a practicing clinician, or a health system administrator developing practice standards. In outlining standards for patient-centered research, the Patient Centered Outcomes Research Institute encourages providing information supporting the selection of outcomes as clinically meaningful, such as input from patients and their families. Neonatal Comparative Effectiveness 839 Deciding which stakeholder’s perspective drives outcome selection makes a difference in study design. For example, leaders of a NICU wishing for their hospital to compare favorably in outcomes reporting across the Vermont Oxford Network or their multi-unit practice group may be interested in strategies to reduce their unit’s rate of BPD. However, families may be less concerned about whether their infant requires oxygen at 36 weeks’ postmenstrual age, versus whether their infant requires home oxygen at discharge or requires rehospitalization after leaving the NICU. When a reduction in a proximal morbidity such as BPD results in a better long-term health outcome, all stakeholders are mutually satisfied. However, a recent publication from the SUPPORT trial highlights the fact that proximal and distal outcomes are not necessarily equivalent: The study found no difference in the primary composite outcome of death or bronchopulmonary dysplasia,