130 - A Comparison of Clinical and Administrative Methods for Identifying Children with Medical Complexity

Poster Number: 130
Publication Number: 130.304

Ingrid A. Larson, Children's Mercy - Kansas, Leawood, KS, United States; Jeffrey D. Colvin, Children's Mercy Hospitals and Clinics, Kansas City, MO, United States; Rupal Gupta, Children's Mercy Hospitals and Clinics, Leawood, KS, United States; S Margaret Wright, Children's Mercy Hospitals and Clinics, Kansas City, MO, United States; Courtney M. Winterer, Children’s Mercy Kansas City, Kansas City, MO, United States; Cristy Toburen, Children's Mercy Hospitals and Clinics, Shawnee, KS, United States; Kristi Williams, Children's Mercy Hospitals and Clinics, Kansas City, MO, United States; Emily J. Goodwin, Children's Mercy Kansas City, Kansas City, MO, United States; Ryan M. Northup, Children's Mercy Hospitals and Clinics, Kansas City, MO, United States; Isabella Zaniletti, Children's Hospitals Association, Lenexa, KS, United States

Background: Children with medical complexity (CMC) represent a small portion of all children but account for over a third of pediatric health care costs. CMC often receive primary care in specialized CMC clinics, many of which use the HOME instrument to determine which patients qualify (Figure 1). Researchers commonly have used the ICD-10 algorithm for “complex chronic conditions” (CCC) to identify CMC in large administrative healthcare datasets. No previous research has compared HOME scores and CCCs for identifying CMC.

Objective: To determine the accuracy of the HOME instrument in identifying CMC by comparing it to CCCs and to modify it to improve its accuracy.

Design/Methods: In the PHIS healthcare database, we randomly selected 300 primary care patients at a children’s hospital. We used PHIS’s ICD-10 algorithms to stratify the sample to produce 3 patient groups: no chronic condition, non-complex chronic conditions, and CCCs. HOME scoring was completed for each patient. The study population was categorized into tertiles based on the HOME score: highest, middle, and lowest scores. Logistic regression was performed on 80% of the study population to produce odds ratio of having ≥1 CCC and ≥2 CCCs. To improve HOME accuracy, the scoring weights were reweighted proportionately to each sub-scale’s logistic regression coefficient. The logistic regressions were repeated using the re-weighted scales. Finally, we determined the accuracy of the original and reweighted HOME instruments for identifying patients with ≥1 CCC and ≥2 CCCs: we calculated the area under the curve (AUC) using the remaining 20% of the study population. AUC >0.80 was considered “excellent”.

Results: Figure 2 displays the study population characteristics. Using the original HOME score, patients in the highest score tertile had 25 times the odds of having ≥1 CCCs (OR 25.2 [95%CI: 11.6, 54.6]) and 27 times the odds of ≥2 CCCs (OR 27.3 [95%CI: 8.0, 92.4)]. Using those coefficients, the HOME was re-weighted (Figure 1). Using the re-weighted HOME, patients with scores in highest tertile had 19 times the odds of having ≥1< CCC and ≥2 CCCs (OR 19.4 [95%CI: 8.6, 43.9]). Figure 3 displays the AUCs calculated on 20% of the study population: The AUCs for the original HOME were 0.81 for ≥1 CCC and 0.82 for ≥2 CCCs. The re-weighted AUCs were similar.Conclusion(s): The accuracy of the HOME instrument is “excellent” at predicting patients with ≥1 CCC and ≥2 CCCs. Re-weighting the HOME instrument does not improve its performance.
Figure 1. HOME instrument* with original and re-weighted scoring* This tool is adapted with permission of the Center for Medical Home Improvement’s “Exeter Pediatric Associates ‘HOMES’ Complexity Index.” The adaptation used in this study did not include the Social (“S”) subscale.
Figure 2. Demographic and clinical characteristics of the study population