Abstract
<title>Abstract</title> <p> <bold>Objectives</bold> To develop and describe a composite proxy Case Mix Index (CMI) for Indian hospitals operating without Diagnosis-Related Group (DRG) coding infrastructure, and to present a protocol for its prospective validation. <bold>Design:</bold> Model development and planned prospective validation study, reported in accordance with the Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis (TRIPOD) guidelines and the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement. <bold>Setting:</bold> Aster Medcity, a 670-bed quaternary-care private hospital in Kochi, Kerala, India. The validation phase will extend to 3–5 additional tertiary hospitals across India. <bold>Participants:</bold> The preliminary developmental sample comprises 30 consecutive inpatient episodes across 10 specialties (April–June 2025). The planned validation cohort will include ≥ 600 consecutive discharges across ≥ 15 specialties over 12 months. <bold>Primary and secondary outcome measures:</bold> The primary outcome is construct validity of the Proxy CMI, assessed through Spearman rank correlation with direct episode cost (target r ≥ 0.60) and known-group comparison between intensive care unit (ICU) and non-ICU episodes. Secondary outcomes include criterion validity (intraclass correlation coefficient [ICC] ≥ 0.70 with cost index), predictive validity for in-hospital mortality (area under the receiver operating characteristic curve [AUROC] ≥ 0.70), and inter-site reproducibility. <bold>Results</bold> In the preliminary 30-episode sample, the mean Proxy CMI was 0.94 (SD 0.31, range 0.46–1.53). CMI varied across specialties in a pattern consistent with expected clinical complexity: Critical Care (1.36), Infectious Diseases (1.18), and Cardiac Surgery (1.15) ranked highest, while Internal Medicine (0.61) ranked lowest. The Spearman correlation between Proxy CMI and direct cost was 0.76 (p < 0.001). Mean CMI was significantly higher in deceased patients (1.36) than in survivors (0.89, Mann-Whitney U = 8, p = 0.025). <bold>Conclusions</bold> The five-component Proxy CMI demonstrates promising face and construct validity in a preliminary Indian hospital sample. The model uses routinely collected administrative data, requires no DRG infrastructure, and can be implemented in a standard spreadsheet. Formal validation in a larger, multi-centric cohort is warranted and planned. </p>