Quantum Neural Networks for Propensity Score Estimation and Survival Analysis in Observational Biomedical Studies

Vojtěch Novák, Ivan Zelinka, Lenka Přibylová, Lubomír Martínek · 2025

Read Quantum Neural Networks for Propensity Score Estimation and Survival Analysis in Observational Biomedical Studies by Vojtěch Novák, Ivan Zelinka, Lenka Přibylová, Lubomír Martínek (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2506-19973.txt.

Introduction

Vojtech Novak and co-authors test quantum neural networks for medical data work in a 2025 paper. Their test case is a group of more than a thousand patients treated for bowel cancer at a hospital in Ostrava between 2001 and 2009.

The models score each patient by how likely they were to get one of two surgery types, which helps strip bias from survival checks. The team used four key traits and tested the models under exact, sampled, and noisy fake hardware runs. The quantum models beat standard methods on small samples. After matching and weighting, survival did not differ by surgery type, which suggests early gaps came from bias.

Catalog record

CREATORVojtěch Novák, Ivan Zelinka, Lenka Přibylová, Lubomír Martínek
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE93 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2506-19973

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Quantum Neural Networks for Propensity Score Estimation and Survival Analysis in Observational Biomedical Studies Preprint: APS/123-QED Vojtěch Novák, Ivan Zelinka, Lenka Přibylová, Lubomír Martínek Abstract This study investigates the application of quantum neural networks (QNNs) for propensity score estimation to address selection bias in comparing survival outcomes between laparoscopic and open surgical techniques in a cohort of 1177 colorectal carcinoma patients treated at University Hospital Ostrava (2001–2009). Using a dataset with 77 variables, including patient demographics and tumor characteristics, we developed QNN-based propensity score models focusing on four key covariates (Age, Sex, Stage, BMI). The QNN architecture employed a linear ZFeatureMap for data encoding, a SummedPaulis operator for predictions, and the Covariance Matrix Adaptation Evolution Strategy (CMA-ES) for robust, gradient-free optimization in noisy quantum environments. Variance regularization was integrated to mitigate quantum measurement noise, with simulations conducted under exact, sampling (1024 shots), and noisy hardware (FakeManhattanV2) conditions. QNNs, particularly with simulated hardware noise, outperformed classical logistic regression and gradient boosted machines in small samples (AUC up to 0.750 for n=100), with noise modeling enhancing predictive stability. Propensity score matching and weighting, optimized via genetic matching and matching weights, achieved covariate balance with standardized mean differences of 0.0849 and 0.0869, respectively. Survival analyses using Kaplan-Meier estimation, Cox proportional hazards, and Aalen additive regression revealed no significant survival differences post-adjustment (p-values 0.287–0.851), indicating confounding bias in unadjusted outcomes. These results highlight QNNs’ potential, enhanced by CMA-ES and noise-aware strategies, to improve causal inference in biomedical research, particularly for small-sample, high-dimensional datasets. I Introduction Colorectal cancer (CRC) ranks among the most prevalent and lethal malignancies worldwide, with the Czech Republic exhibiting some of the highest incidence rates. Despite its treatability in early stages, late diagnoses and advanced disease stages contribute significantly to its morbidity and mortality burden. Surgical resection remains the cornerstone of curative treatment, with laparoscopic and open surgical techniques representing the primary approaches. Laparoscopic surgery offers advantages such as reduced blood loss, faster recovery, and lower infection risk, but its long-term survival benefits compared to open surgery remain uncertain, particularly in observational studies where treatment assignment is influenced by patient characteristics like tumor stage and comorbidities. Addressing this selection bias is critical for unbiased treatment effect estimation, necessitating advanced statistical and computational methods. Survival analysis provides a robust framework for evaluating time-to-event outcomes, such as survival time, while handling censoring and covariate effects Klein and Moeschberger 2003 . However, observational studies, unlike randomized controlled trials, face challenges from non-random treatment allocation, which can confound outcome comparisons. Propensity score methods, including matching and weighting, mitigate this bias by balancing covariate distributions between treatment groups, enabling causal inference Rosenbaum and Rubin 1983 . Traditional approaches like logistic regression and gradient boosted models (GBM) estimate propensity scores but struggle with high-dimensional biomedical data and complex covariate interactions McCaffrey et al. 2004 , Přibylová et al. 2023 . Quantum machine learning (QML), particularly quantum neural networks (QNNs), offers a promising alternative by leveraging quantum computing’s unique properties—such as superposition and entanglement—to enhance optimization and data processing in high-dimensional spaces Biamonte et al. 2017 . In the noisy intermediate-scale quantum (NISQ) era, hybrid quantum-classical QNNs, built on parameterized quantum circuits (PQCs), provide a feasible approach for tasks like propensity score estimation, potentially outperforming classical methods in small-sample or complex datasets Schuld and Killoran 2019 . This study investigates the application of QNNs to propensity score estimation for comparing survival outcomes between laparoscopic and open surgery in a cohort of 1177 CRC patients treated at University Hospital Ostrava from 2001 to 2009. The dataset, comprising 77 variables including patient demographics, tumor characteristics, and surgical outcomes, exhibits non-random treatment assignment, necessitating robust bias correction. We compare QNN-based propensity score models, implemented under exact, sampling, and noisy hardware simulations, against classical logistic regression and GBM across sample sizes (n=100, 500, and full dataset) to assess their performance in small-sample and high-dimensional settings. Propensity scores are applied through matching and weighting, with covariate balance evaluated via standardized mean differences (SMDs) and survival outcomes analyzed using Kaplan-Meier estimation, Cox proportional hazards models, and Aalen additive regression Klein and Moeschberger 2003 . This work aims to demonstrate QNNs’ potential to enhance causal inference in biomedical research, offering insights into surgical technique efficacy and advancing the integration of quantum computing in healthcare analytics. II Introduction to the Medical Problem Colorectal cancer (CRC) is a malignant disease of the digestive system affecting the colon and rectum. It is among the most common cancers worldwide in terms of both incidence and mortality Bray et al. 2018 . For several years, the Czech Republic has ranked among the countries with the highest rates of this malignancy Dusek et al. 2014 . Although CRC is highly treatable in its early stages, its rising incidence and the frequent diagnosis at advanced stages impose a significant burden of morbidity and mortality on the population. Therefore, public and professional awareness is crucial for prevention and early detection Winawer et al. 2003 . The causes of CRC can be divided into external (exogenous) and internal (endogenous) risk factors. Exogenous factors primarily relate to diet and lifestyle habits, such as the consumption of red meat and animal fats, particularly when fried, roasted, or smoked, which can produce carcinogenic compounds Zhao et al. 2017 . Excessive alcohol consumption and a diet lacking in fiber from fruits and vegetables, calcium, vitamins C and D, and folic acid are also considered risk factors Slattery et al. 1998 . Endogenous factors, which account for about 15-20% of cases, include hereditary and familial predispositions Lynch and De la Chapelle 2003 . However, the most common form, accounting for 80-85% of all cases, is sporadic CRC, which typically develops from benign adenomatous polyps Ahadova et al. 2018 . The single most significant endogenous risk factor for sporadic CRC is age, with a notable increase in incidence after the age of fifty Haggar and Boushey 2009 . In their initial stages, colorectal tumors are often asymptomatic. The only sign may be occult (hidden) bleeding in the stool, which is not visible to the naked eye but can be detected through testing Simon 1984 . As these early-stage tumors cause few noticeable problems, patients may not seek attention, delaying the start of treatment. Often, the first indication of an advanced tumor is an acute, life-threatening condition such as a bowel obstruction, which manifests with abdominal pain, cessation of gas and stool, diarrhea, vomiting, and visible bleeding Deans et al. 1994 . If an obstruction develops, surgery becomes essential to prevent a breach of the intestinal barrier, which could lead to shock and death. For malignant tumors, surgery is the primary and often only curative treatment option Cunningham et al. 2010 . Diagnosis typically begins with a colonoscopy, which allows for direct visualization of the colon lining and the collection of a tissue sample (biopsy) for analysis Winawer et al. 2003 . For rectal tumors, a rectoscopy is performed. Further diagnostic methods are used to determine the extent of the disease. Imaging techniques such as computed tomography (CT), magnetic resonance imaging (MRI) or positron emission tomography/CT (PET/CT) are employed to assess the extent of disease, lymph node involvement and the presence of metastases, while a chest X-ray is used to check

Read the complete text