Matrix factorization for the reconstruction of cervical cancer screening histories and prediction of future screening results.

BACKGROUND: Mass screening programs for cervical cancer prevention in the Nordic countries have strongly reduced cancer incidence and mortality at the population level. An alternative to the current mass screening is a more personalised screening strategy adapting the recommendations to each individual. However, this necessitates reliable risk prediction models accounting for disease dynamics and individual data. Herein we propose a novel matrix factorisation framework to classify females by the time-varying risk of being diagnosed with cervical cancer. We cast the problem as a time-series prediction model where the data from females in the Norwegian screening population are represented as sparse vectors in time and then combined into a single matrix. Using novel temporal regularisation and discrepancy terms for the cervical cancer screening context, we reconstruct complete screening profiles from this scarce matrix and use these to predict the next exam results indicating the risk of cervical cancer. The algorithm is validated on both synthetic and registry screening data by measuring the probability of agreement (PoA) between Kaplan-Meier estimates. RESULTS: In numerical experiments on synthetic data, we demonstrate that the novel regularisation and discrepancy term can improve the data reconstruction ability as well as prediction performance over varying data scarcity. Using a hold-out set of screening data, we compare several numerical models and find that the proposed framework attains the strongest PoA. We observe strong correlations between the empirical survival curves from our method and the hold-out data, and evaluate the ability of our framework to predict the females' next results for up to five years ahead in time using only their current screening histories as input. CONCLUSIONS: We have proposed a matrix factorization model for predicting future screening results and evaluated its performance in a female cohort to demonstrate the potential for developing prediction models for more personalized cervical cancer screening.

Towards a data-driven system for personalized cervical cancer risk stratification.

Mass-screening programs for cervical cancer prevention in the Nordic countries have been effective in reducing cancer incidence and mortality at the population level. Women who have been regularly diagnosed with normal screening exams represent a sub-population with a low risk of disease and distinctive screening strategies which avoid over-screening while identifying those with high-grade lesions are needed to improve the existing one-size-fits-all approach. Machine learning methods for more personalized cervical cancer risk estimation may be of great utility to screening programs shifting to more targeted screening. However, deriving personalized risk prediction models is challenging as effective screening has made cervical cancer rare and the exam results are strongly skewed towards normal. Moreover, changes in female lifestyle and screening habits over time can cause a non-stationary data distribution. In this paper, we treat cervical cancer risk prediction as a longitudinal forecasting problem. We define risk estimators by extending existing frameworks developed on cervical cancer screening data to incremental learning for longitudinal risk predictions and compare these estimators to machine learning methods popular in biomedical applications. As input to the prediction models, we utilize all the available data from the individual screening histories.Using data from the Cancer Registry of Norway, we find in numerical experiments that the models are strongly biased towards normal results due to imbalanced data. To identify females at risk of cancer development, we adapt an imbalanced classification strategy to non-stationary data. Using this strategy, we estimate the absolute risk from longitudinal model predictions and a hold-out set of screening data. Comparing absolute risk curves demonstrate that prediction models can closely reflect the absolute risk observed in the hold-out set. Such models have great potential for improving cervical cancer risk stratification for more personalized screening recommendations.