IMPROVING IDENTIFICATION OF PATIENTS ELIGIBLE FOR ENDOGENOUS CUSHING DISEASE TREATMENT THROUGH CLINICAL EXPERTISE AND MACHINE LEARNING
Author(s)
Nicola Lazzarini, PhD1, Kunal Puri, MD2, Ban Tawfik, BSc1, Tom Wijnands, MSc3, Thomas Hupp, PhD4, Andrea Seltmann, Dipl.Soz.5.
1IQVIA, London, United Kingdom, 2IQVIA, Bangalore, India, 3IQVIA, Amsterdam-Zuidoost, Netherlands, 4IQVIA, Frankfurt, Germany, 5IQVIA, Berlin, Germany.
1IQVIA, London, United Kingdom, 2IQVIA, Bangalore, India, 3IQVIA, Amsterdam-Zuidoost, Netherlands, 4IQVIA, Frankfurt, Germany, 5IQVIA, Berlin, Germany.
OBJECTIVES: Endogenous Cushing Disease is a rare disorder, affecting 40-80 per million in Europe. This study aims to use clinical expertise and machine learning to identify patients best suited for novel treatments, including those on other medications or with post-surgical symptoms.
METHODS: We used a hybrid approach combining clinical expertise and machine learning (ML). A treatment eligibility ML model was developed using IQVIA prescription data (LRx) and trained on 1,272 patients considered probable endogenous Cushing Disease cases, in addition to 299 patients treated with cortisol synthesis inhibitor who served as the target cohort. For each patient, treatment journey information was derived from longitudinal prescription records and used as input features for the model to learn treatment patterns associated with the target therapy.
RESULTS: The ML model achieved a 72% recall rate in correctly identifying target patients. The predictive features used by the model matched clinical expectations, with treatments such as pasireotide, spironolactone and mitotane being key contributors in the model predictions. Of the 1,272 probable Cushing Disease cases identified by clinical experts, the model prioritized 321 patients as most likely to switch to targeted treatment if given the option. Furthermore, when prospectively reviewing patients who initiated Cushing treatment within three months, 30% of them were captured by the clinical expert selection rules with a 10% also flagged as priority by the model, indicating strong performance for a rare disease. Differences in the regional geographic distribution of existing target patients and newly identified eligible patients were also observed, illustrating the potential to uncover untapped treatment opportunities.
CONCLUSIONS: Combining clinical expertise with machine learning enables a more precise and actionable identification of eligible patients in real world data, helping to focus treatment opportunities in rare and hard-to-find populations. This approach can help uncover untapped opportunities and support more targeted patient activation strategies.
METHODS: We used a hybrid approach combining clinical expertise and machine learning (ML). A treatment eligibility ML model was developed using IQVIA prescription data (LRx) and trained on 1,272 patients considered probable endogenous Cushing Disease cases, in addition to 299 patients treated with cortisol synthesis inhibitor who served as the target cohort. For each patient, treatment journey information was derived from longitudinal prescription records and used as input features for the model to learn treatment patterns associated with the target therapy.
RESULTS: The ML model achieved a 72% recall rate in correctly identifying target patients. The predictive features used by the model matched clinical expectations, with treatments such as pasireotide, spironolactone and mitotane being key contributors in the model predictions. Of the 1,272 probable Cushing Disease cases identified by clinical experts, the model prioritized 321 patients as most likely to switch to targeted treatment if given the option. Furthermore, when prospectively reviewing patients who initiated Cushing treatment within three months, 30% of them were captured by the clinical expert selection rules with a 10% also flagged as priority by the model, indicating strong performance for a rare disease. Differences in the regional geographic distribution of existing target patients and newly identified eligible patients were also observed, illustrating the potential to uncover untapped treatment opportunities.
CONCLUSIONS: Combining clinical expertise with machine learning enables a more precise and actionable identification of eligible patients in real world data, helping to focus treatment opportunities in rare and hard-to-find populations. This approach can help uncover untapped opportunities and support more targeted patient activation strategies.
Conference/Value in Health Info
2026-11, ISPOR Europe 2026, Vienna, Austria
Value in Health, Volume 29, Issue 12S
Code
MSR94
Topic
Methodological & Statistical Research, Real World Data & Information Systems, Study Approaches
Topic Subcategory
Artificial Intelligence, Machine Learning, Predictive Analytics
Disease
Diabetes/Endocrine/Metabolic Disorders (including obesity), No Additional Disease & Conditions/Specialized Treatment Areas, Rare & Orphan Diseases