Benefits and Cost-Effectiveness of a Pediatric Hospitalist Service : a Stepped Wedge Quality Improvement Trial

Background Inpatient hospital services constitute the largest share (29%) of total national health care spending. Hospitalizations for children cost $37 billion annually. The traditional model, in which inpatient care has been provided by primary care pediatricians is being replaced by a new model of pediatric hospitalists whose primary or exclusive professional focus is the care of hospitalized patients. This shift has been driven by the increasing illness severity and complexity of hospitalized children, the challenges in managing these children by office-based pediatricians, and the increasing demand for higher quality, and more efficient and cost-effective hospital care. Although the pediatric hospitalist model system is being widely disseminated, its value in improving clinical outcomes or controlling health care costs remains to be demonstrated. The findings in studies of pediatric or adult hospitalists have been mixed, and almost all studies have been of low quality, mostly retrospective before and after comparisons subject to a variety of biases. With the necessity to restrain health care spending, the funding required from Medicaid and other 3rd-party payers to develop and sustain high-quality pediatric hospitalists services may not be forthcoming unless they are shown to improve outcomes with acceptable or reduced costs. The investigators propose to conduct an evaluation of the cost-effectiveness of pediatric hospitalist care assessed as described below in a stepped wedge QI trial. The pediatric hospitalist program will be initiated by the Department of Pediatrics with Medical School support and is planned for stepwise implementation at Children's Memorial Hermann Hospital (CMHH) starting in 2016. Patient and Methods Aims. To assess in the most rigorous feasible manner whether the pediatric hospitalist services soon to be implemented at CMHH is cost-effective--expressed as net cost or savings per hospital day prevented (from the health system perspective) -as well as beneficial in improving clinical outcomes and parent, nurse, and physician satisfaction relative to that with the general pediatrics service. Trial Hypothesis: The hospitalist program will be cost-effective by either reducing hospital days without increasing health system costs, reducing health system costs without increasing hospital days, or reducing both. Design: Stepped Wedge Quality Improvement Trial. Rationale. It would not be desirable or feasible to use a conventional parallel group randomized trial for at least two reasons: 1) With new hires, the number of hospitalists will increase substantially over the study period, and a fixed allocation of patients (e.g., the usual 1:1 ratio) to the hospitalist and general pediatrics services throughout the study would not efficiently use these faculty; 2) Random assignment of patients would be difficult to implement, might result in unacceptable imbalances in the number of patients assigned to the 4 different hospitalist or general pediatric teams, and would intrude on current routines and likely be resisted by the hospital administration. Stepped-wedge trials involve sequential roll-out of an intervention to groups- in this care, 4 hospital services, over a number of time periods. At the end of the entire study all groups will have received the intervention. Each serves as its own control in a before/after comparison and also as control for other groups in comparison of groups across time. Thus, this design allows both between group and within-group analyses. Moreover, it also allows flexibility in deciding when to initiate a stepwise increase in patients served and in augmenting the intervention strategy based upon problems and solutions identified over time. This design is particularly appropriate in implementing interventions like a hospitalist program for which there is likely clinical benefit with minimal risk, and it is not feasible for logistical or financial reasons to either randomize or deliver the intervention simultaneously to all participants. To this end, the investigators will use a stepped-wedge QI trial design to serially introduce the hospitalist program in all 4 inpatient services at CMHH currently staffed by primary care pediatricians from the General Pediatric Division. The researchers have chosen to use this design because it allows partial adjustment for any important temporal changes that may occur unrelated to the intervention, and facilitates a more rigorous assessment of treatment effects than the usual pre- and post-intervention comparisons, while also allowing the investigators to start the project quickly with the available faculty and to focus on optimizing hospitalist care in each service successively. Patients. Children 18 years and younger admitted to the 4 pediatric inpatient services (including those on observation status) at CMHH. These services include inpatient ward and intermediate care beds. (Patients who are first admitted to the pediatric intensive care unit \[ICU\] will not be considered enrolled in the study until they are transferred to the pediatric service.) To adjust for any important chance differences between services in baseline patient risk, study personnel blinded to the service will categorize each child as high risk or very high risk based on findings present on admission to a general pediatrics or hospitalist service. This designation will be based on predefined criteria that are easily applied on admission and related to the risk of prolonged or expensive hospitalizations (e.g. transfer from the pediatric ICU, admission to the IMU, mechanical ventilation, with do-not-resuscitate order, or scheduled for surgery). Treatment Allocation. Currently, the 4 pediatric inpatient services each admit patients every other day; the 2 admitting services each accept every other patient without regard to diagnosis. These procedures will not be altered by the study to avoid disrupting current hospital routines while still allowing quasi-randomized assignment to the different services that is unaffected by the child's prior history or his/her diagnoses and condition on admission. The order in which the hospitalist services will be implemented will be randomly determined. General Pediatric Services at CMHH. The attending pediatrician on each service conducts daily morning rounds on weekdays, supervises the students and house staff and is available to the residents by phone or in person during the day. On weeknights, an on-call pediatrician is available to the residents by phone. On weekends two on-call pediatricians make rounds, supervise the residents, and take call from home for the 4 services. This approach on the General Pediatric Services will not be changed materially as hospitalists assume responsibility for one or more of the 4 services. Hospitalist Services at CMHH. The hospitalist services at MHH will be phased in according to the number of hospitalists who are appointed to the new Pediatric Hospitalist Division. As currently planned hospitalist services will involve an in-house hospitalist at all hours. The attending hospitalist on each hospitalist service will make morning rounds on weekdays and be present supervising the residents or providing care throughout the day. A different hospitalist will cover at night. On weekends, one hospitalist will cover up to 2 services each day. Outcomes: Primary: cost-effectiveness, expressed as net cost or savings per hospital day prevented assessed from the health system perspective. Secondary: Hospital days (including 30-day readmissions), severe deterioration (transfer to the pediatric ICU, intubation, unexpected surgery, or death) days with diversion of patients from the ED, pediatric ICU, or IMU), parent satisfaction at discharge, nurse and resident satisfaction assessed once a year, and costs expressed from the healthcare system perspective. Sample Size. The sample size won't and can't be predefined. The most important problem that a predefined sample size is used to avoid - bias in deciding when to stop enrollment -is unlikely to be a problem when the decision to stop enrollment will be based on factors (in this case, the recruitment of hospitalists) that are unlikely to hinge on the accrued results. Another problem that a fixed sample size is intended to avoid is imprecision of the effect size identified. For this study, the precision of this estimate will be limited by the numbers of children who are admitted at the 4 services over the 2-3 year study period. This is likely to be the largest feasible sample size that can be studied, as \>8,000 admits are expected in 3 yrs. Statistical Analyses. All analyses will be conducted at the individual patient level using a generalized linear mixed model (GLMM). Separate GLMMs will be used for outcome. GLMMs account for the stepped wedge design, individual level covariates, and service level covariates, and allow each service to have different number of patients at each time point. For all secondary outcomes, the investigators will investigate the best fitting distributional form (i.e., Normal, gamma) to estimate relative risks (RRs) for the intervention effect. Each model will include treatment group (intervention/control), time (5 planned periods), season, age, ethnicity, insurance status, baseline patient risk, time of hospital admission (day on weekdays: 8 am - 5 pm; weekday night; weekend) as covariates, and a random effect for child (to account for within subject correlation due to possible repeat admissions by the same child). Estimates of RRs and 95% credible intervals will be reported with posterior probabilities of benefit and cost-effectiveness. Prior distributions for all covariates will be neutral but excluding implausible large treatment effects \~N(0, SD=0.35). Interim analyses will be performed 6 months after hospitalists become responsible for the 2nd and 3rd services to verify that the hospitalist program does not unexpectedly worsen outcome or increase costs before expanding the program to all 4 services. Otherwise the study will continue as planned with final analyses performed on all patients admitted by 6 months after the hospitalists assume responsibility for the 4th and final service. The investigators will conclude that hospitalist program is cost effective if Bayesian analyses indicate a 90% or higher probability that it reduces hospital days and/or reduces costs. Cost-Effectiveness Analyses. Hospital days will be computed as the number of hospital days after admission to any of the four services (not including any days spent in other hospitals or the pediatric ICU before transfer to one of these services) and any subsequent readmission occurring within 30 days of discharge. These data will be obtained by searching the claims data for the 13 Memorial Hermann Health System hospitals. Hospitalizations outside this system will be identified for high-risk chronically ill children as part of a separate ongoing QI study. If as expected, parent satisfaction is increased and/or arrangements for care after discharge is greater with hospitalist care, any rehospitalizations not identified would be expected to be less with hospitalist care, a factor that would result in a conservative estimate of any reduction in total hospital days with such care. The incremental costs of hospitalist care will be related to prevented hospital days from a healthcare system perspective following state-of-the-art guidelines. Hospital costs after admission to one of the 4 services (excluding any prior costs incurred in the emergency room \[ER\] or pediatric intensive care unit \[ICU\]) will be estimated using claims-based hospital charges multiplied by department-specific cost-to-charge ratios specified in the hospital's annual Medicare Cost Report. The costs will be inflated to 2018 U.S. dollars based on the Consumer Price Index for medical services. Physician costs will be estimated using a relative value unit (RVU)-based method. The resource-based RVU for each physician service provided during hospital days will be obtained from the medical school billing system and multiplied by the Medicare conversion factor. Costs for hospitalist care will be augmented to include any salary differentials and/or supplements received by hospitalists for taking more night shifts. Differences in cost and hospital days between treatment groups will be assessed using generalized linear models with log-link and gamma distribution. The model will include treatment group (intervention/control), time (5 periods), season, age, ethnicity, insurance status, baseline patient risk , time of hospital admission (day on weekdays: 8 am - 5 pm; weekday night; weekend) as covariates, and will account for correlations within patient. The incremental cost (savings) of hospitalist care will be computed by subtracting the mean cost per patient in this group from the mean cost per patient in the general pediatrics group. The incremental cost-effectiveness ratio will be computed by dividing the incremental costs (savings) by the hospital days avoided. The investigators will also assess the robustness of the results in sensitivity analyses by assessing the joint influence of hospitalists care on both cost and effectiveness using the net monetary benefit approach. Secondary Economic Analysis. Analyses from the medical school perspective will evaluate the effect of the hospitalist intervention relative to usual care on the difference between revenues and costs (i.e. net gain or loss) for inpatient care, excluding research-related costs.