A Multi-Indicator Concentration Index for Primary-Care Reimbursement Analytics: Evidence from the Albanian Health-Insurance System

Vollständiger Abstract

Worum geht es in dieser Arbeit?

Background: Public payers must regularly decide which prescribers to look at more closely, with records covering thousands of doctors per quarter and review capacity for only a fraction. How that choice is made bears on equitable access to reimbursed medicines as well as on public funds. Methods: We propose a multi-indicator concentration index scoring each prescriber 0–100 as the sum of nine weighted sub-scores, each measuring the doctor’s deviation from their primary-care centre median on one indicator from the payer’s routine quarterly report; eight are computable from a single quarter. Every score decomposes into named components. The index measures the concentration of indicators above the peer profile, not an estimated probability of misconduct. It is a screening tool for prioritising documentary review. We demonstrate it on one quarter of anonymised data from 335 doctors in 30 Albanian primary-care centres. Results: The eight active sub-scores are mostly weakly correlated (mean |r|=0.28), the two exceptions each pairing a case count with its financial effect. Rankings are robust to alternative weights (Spearman ρ between 0.93 and 0.98), to collapsing those pairs (ρ≥0.98), and to pooling the smallest centres against a shared comparator (ρ=0.95). Against three machine-learning anomaly detectors and two simpler screens, the index selects a different top cohort: doctors whose spending is only moderately raised, but whose new-case and therapy-change counts exceed three times the centre median. In an internal concordance check, two co-author reviewers blinded to the score rated 60 cases sampled partly by index rank; they agreed with each other (Cohen’s κ=0.87, 95% CI 0.70 to 1.00), and the index’s ranking against their consensus gives an AUC of 0.88 (95% CI 0.62 to 0.97). Conclusions: The index gives clinical-administrative reviewers a reproducible, explainable priority order consistent with senior expert judgement on the same indicators. Because no review outcomes are linked to the records and the window is a single quarter, it is supported as a prioritisation tool, not a validated predictor of inappropriate reimbursement; multi-quarter and outcome-linked evaluation are the next steps.