From one-size-fits-all to tailored care: navigating trade-offs in efficiency, equity, and interpretability in healthcare delivery
File(s)
Author(s)
Mohammadi, Narges
Type
Thesis
Abstract
Healthcare delivery increasingly requires balancing efficiency, equity, and interpretability as systems move from standardized protocols toward more tailored, patient-centered approaches. This thesis investigates how analytics and optimization can guide the design of clinical policies that achieve these trade-offs while remaining computationally and practically feasible.
The first part of this work addresses hearing loss screening in cystic fibrosis (CF) patients, where repeated use of aminoglycoside antibiotics increases the risk of ototoxicity. A Partially Observable Markov Decision Process (POMDP) framework is developed to integrate mobile and formal audiometry, optimizing screening frequency and modality. A novel algorithm enables systematic sensitivity analysis over willingness-to-pay (WTP) thresholds, allowing policymakers to identify cost-effective policies across a wide economic spectrum.
The second part focuses on stratified cancer surveillance. We develop a multi-model constrained POMDP framework that transforms optimization under semi-Markov disease dynamics into a tractable single POMDP model. By combining biomarker-based (ctDNA) testing with imaging, we generate surveillance schedules that align with guidelines while accounting for resource constraints. Results show that optimized strategies can reduce false positives by an order of magnitude, maintain timely treatment, and deliver substantial economic value, with tailored policies for meaningful subpopulations.
Finally, the thesis addresses the broader tension between one-size-fits-all and personalized medicine. Using tree-based partitioning and proxy objectives, it develops tractable stratification strategies that balance complexity, efficiency, and outcome equity. This demonstrates how tailored care can be systematically designed without sacrificing interpretability or equity.
Together, these contributions advance both methodology and practice: introducing efficient algorithms for cost-effectiveness frontier construction, developing a multi-model CPOMDP framework for stratified surveillance, and proposing optimal stratification strategies via decision trees. Beyond healthcare, the frameworks generalize to public policy and operations, where competing priorities must be carefully navigated.
The first part of this work addresses hearing loss screening in cystic fibrosis (CF) patients, where repeated use of aminoglycoside antibiotics increases the risk of ototoxicity. A Partially Observable Markov Decision Process (POMDP) framework is developed to integrate mobile and formal audiometry, optimizing screening frequency and modality. A novel algorithm enables systematic sensitivity analysis over willingness-to-pay (WTP) thresholds, allowing policymakers to identify cost-effective policies across a wide economic spectrum.
The second part focuses on stratified cancer surveillance. We develop a multi-model constrained POMDP framework that transforms optimization under semi-Markov disease dynamics into a tractable single POMDP model. By combining biomarker-based (ctDNA) testing with imaging, we generate surveillance schedules that align with guidelines while accounting for resource constraints. Results show that optimized strategies can reduce false positives by an order of magnitude, maintain timely treatment, and deliver substantial economic value, with tailored policies for meaningful subpopulations.
Finally, the thesis addresses the broader tension between one-size-fits-all and personalized medicine. Using tree-based partitioning and proxy objectives, it develops tractable stratification strategies that balance complexity, efficiency, and outcome equity. This demonstrates how tailored care can be systematically designed without sacrificing interpretability or equity.
Together, these contributions advance both methodology and practice: introducing efficient algorithms for cost-effectiveness frontier construction, developing a multi-model CPOMDP framework for stratified surveillance, and proposing optimal stratification strategies via decision trees. Beyond healthcare, the frameworks generalize to public policy and operations, where competing priorities must be carefully navigated.
Version
Open Access
Date Issued
2025-10-03
Date Awarded
01/11/2025
Advisor
Skandari, MohammadReza
Shah, Anand
Publisher Department
Business School
Publisher Institution
Imperial College London
Qualification Level
Doctoral
Qualification Name
Doctor of Philosophy (PhD)