DISCUSSION
The present descriptive study demonstrated that expenditures for patients with SCD in the CHECK High and Medium utilizer tiers decreased over time. However, the small size of the High utilizer group (n=32) may have reduced the statistical power to detect a true effect. The SCD patients in the Low utilizer tier showed significant rise in expenditure levels over time. The Low utilizer tier group increase in expenditures after the second year of CHECK may have been due to receiving appropriate, but formerly unused, services introduced by the CHECK CHWs.
Chronic health conditions effect an estimated 25% of children and adolescents in the United States and account for a disproportionate share of pediatric healthcare utilization and expenditures, with the majority of the $110 billion annual expenditures attributed to hospital care34. Similarly, while comprising a small proportion of children and adolescents with chronic diseases, those with SCD have a disproportionately large effect on health care expenditures41.
In developing the CHECK program, the hope was that a comprehensive health care coordination program would reduce health care expenditures. We did not expect this program would impact all patients equally. While the overall analysis of the CHECK program showed no change in costs related to the program33, this exploration specific to the children with SCD demonstrated a likely impact on healthcare costs for SCD children.
The comparisons between Low and High, Medium and High utilization risk groups support the expectation that the odds of zero expenditures were significantly different from Baseline Year in the predicted direction. We can speculate that care coordination was associated with some of these changes, but the study was designed as a demonstration project and has limitations (discussed below) that preclude definitive conclusion. A null hypothesis that High and Low utilization are simply exhibiting regression to the mean cannot be completely excluded. The high odds ratio of 240.54 for the comparison at Baseline year is likely due to cells with few patients. Finally, the bimodal distributions of utilization for all the risk groups (Figure 1) show striking similarity in Year 2. This longitudinal pattern suggests that one year of utilization data was not enough to stratification patients for utilization.
A longitudinal study of 101 children with SCD in Houston in 2011-2013 found that 25% of parents reported receiving care coordination help and another 20% reported a need for extra care coordination42. However, multivariate regression did not detect an association between care coordination and acute care utilization. The Houston findings are consistent with the stratification we used. Together, the CHECK program data and Houston data suggest that much a larger sample size and a rigorous study design might be required to determine a benefit of care coordination for children with SCD. This is difficult to do because SCD is a rare disease.
The longitudinal pattern of acute care utilization highlights that some SCD patients have complications that vary from year to year. These data suggest that future stratification by utilization risk groups should use baseline data that spans at least 3 years, and the intervention period probably should also span multiple years.
Limitations
This study has several limitations. First, because the CHECK program was designed as a healthcare demonstration project and not as a randomized clinical trial, there is no control group; patients receiving CHECK services were targeted for services for specific reasons. The SCD sample is small, illustrating the small proportion of chronically ill patients with the disorder but also limiting the statistical power necessary for introducing other variables in the analyses. Using these heterogeneous cost data, a sample size estimate for 3 utilization categories with repeated differences in expenditures across time and 2-3 covariates, alpha=.05 and power=.80 would require n=3800 SCD patients in each utilization category to detect a 10% cost difference. The SCD patients in this study were not selected using randomized sampling procedures and outliers were removed, which weakens the ability to generalize findings to the wider SCD population.
Only 52% of SCD patients were fully “engaged” in the CHECK program, meaning they received full CHECK services tailored to individual needs. This means our analysis does include some who may have needed services but did not choose to participate or could not be reached. Some acute hospitalizations in SCD are avoidable (e.g., hospitalization for sickle cell pain that was triggered by bad behavioral choices or failure to fill prescriptions for home pain medications), but other hospitalizations are unavoidable (e.g., fever or Acute chest syndrome). This makes determination of the appropriateness of hospitalizations difficult. New guidelines recommending pediatric use of the disease-modifying therapy hydroxyurea were released in 2014 during the CHECK study period and could have reduced the acute care utilization with or without CHECK services43. CHECK SCD patients received medical services, care coordination, or education from other programs, confounding the coordinated care in CHECK. Most of the CHECK SCD patients were outside of UI Health, receiving a variety of services in other pediatric SCD programs in the Chicago region. The study design does not allow untangling a CHECK treatment effect from these other services. Finally, the longitudinal observational study design leaves open the possibility that the pattern simply represents the phenomenon that data at the extremes of a distribution might show “regression toward the mean” through random changes.
Conclusion
In conclusion, this analysis provides lessons regarding care coordination service needs for children. The heterogeneity of SCD care utilization was revealed in the skewed bimodal distribution of expenditures that emerged in each tier over the three years of data. Stratification showed a trend of reduced expenditures in the High risk and Medium risk groups. The CHECK care coordination intervention was a tailored “precision medicine” approach delivered by CHWs. With this, the care coordination tried to match the intensity of intervention to the multiple modifiers and barriers to SCD acute and chronic care. The benefit of CHECK might have been obscured by low sample size, uneven intensity of care coordination intervention, duration of intervention shorter than the natural variability of SCD expenditures (possibly 3 years), and the inability to distinguish the appropriateness of healthcare utilization for SCD. Future studies of care coordination in SCD should build upon these lessons learned in CHECK: stratify by tiers of healthcare utilization, focus on the highest utilizers, design more than three years of data collection, and add outcomes for appropriateness of healthcare utilization.