Comparison of dynamic updating strategies for clinical prediction models

Dynamic updating of clinical survival prediction models in a changing environment

BACKGROUND

Over time, the performance of clinical prediction models may deteriorate due to changes in clinical management, data quality, disease risk and/or patient mix. Such prediction models must be updated in order to remain useful. In this study, we investigate dynamic model updating of clinical survival prediction models. In contrast to discrete or one-time updating, dynamic updating refers to a repeated process for updating a prediction model with new data. We aim to extend previous research which focused largely on binary outcome prediction models by concentrating on time-to-event outcomes. We were motivated by the rapidly changing environment seen during the COVID-19 pandemic where mortality rates changed over time and new treatments and vaccines were introduced.

METHODS

We illustrate three methods for dynamic model updating: Bayesian dynamic updating, recalibration, and full refitting. We use a simulation study to compare performance in a range of scenarios including changing mortality rates, predictors with low prevalence and the introduction of a new treatment. Next, the updating strategies were applied to a model for predicting 70-day COVID-19-related mortality using patient data from QResearch, an electronic health records database from general practices in the UK.

RESULTS

In simulated scenarios with mortality rates changing over time, all updating methods resulted in better calibration than not updating. Moreover, dynamic updating outperformed ad hoc updating. In the simulation scenario with a new predictor and a small updating dataset, Bayesian updating improved the C-index over not updating and refitting. In the motivating example with a rare outcome, no single updating method offered the best performance.

CONCLUSIONS

We found that a dynamic updating process outperformed one-time discrete updating in the simulations. Bayesian updating offered good performance overall, even in scenarios with new predictors and few events. Intercept recalibration was effective in scenarios with smaller sample size and changing baseline hazard. Refitting performance depended on sample size and produced abrupt changes in hazard ratio estimates between periods.

Automatic correction of performance drift under acquisition shift in medical image classification

Image-based prediction models for disease detection are sensitive to changes in data acquisition such as the replacement of scanner hardware or updates to the image processing software. The resulting differences in image characteristics may lead to drifts in clinically relevant performance metrics which could cause harm in clinical decision making, even for models that generalise in terms of area under the receiver-operating characteristic curve. We propose Unsupervised Prediction Alignment, a generic automatic recalibration method that requires no ground truth annotations and only limited amounts of unlabelled example images from the shifted data distribution. We illustrate the effectiveness of the proposed method to detect and correct performance drift in mammography-based breast cancer screening and on publicly available histopathology data. We show that the proposed method can preserve the expected performance in terms of sensitivity/specificity under various realistic scenarios of image acquisition shift, thus offering an important safeguard for clinical deployment.

Dynamic prediction modeling of postoperative mortality among patients undergoing surgical aortic valve replacement in a statewide cohort over a 12-year period

Objective

Clinical prediction models for surgical aortic valve replacement mortality, are valuable decision tools but are often limited in their ability to account for changes in medical practice, patient selection, and the risk of outcomes over time. Recent research has identified methods to update models as new data accrue, but their effect on model performance has not been rigorously tested.

Methods

The study population included 44,546 adults who underwent an isolated surgical aortic valve replacement from January 1, 1999, to December 31, 2018, statewide in Pennsylvania. After chronologically splitting the data into training and validation sets, we compared calibration, discrimination, and accuracy measures amongst a nonupdating model to 2 methods of model updating: calibration regression and the novel dynamic logistic state space model.

Results

The risk of mortality decreased significantly during the validation period (P < .01) and the nonupdating model demonstrated poor calibration and reduced accuracy over time. Both updating models maintained better calibration (Hosmer-Lemeshow χ2 statistic) than the nonupdating model: nonupdating (156.5), calibration regression (4.9), and dynamic logistic state space model (8.0). Overall accuracy (Brier score) was consistently better across both updating models: dynamic logistic state space model (0.0252), calibration regression (0.0253), and nonupdating (0.0256). Discrimination improved with the dynamic logistic state space model (area under the curve, 0.696) compared with the nonupdating model (area under the curve, 0.685) and calibration regression method (area under the curve, 0.687).

Conclusions

Dynamic model updating can improve model accuracy, discrimination, and calibration. The decision as to which method to use may depend on which measure is most important in each clinical context. Because competing therapies have emerged for valve replacement models, updating may guide clinical decision making.

Methodological guidance for the evaluation and updating of clinical prediction models: a systematic review

BACKGROUND

Clinical prediction models are often not evaluated properly in specific settings or updated, for instance, with information from new markers. These key steps are needed such that models are fit for purpose and remain relevant in the long-term. We aimed to present an overview of methodological guidance for the evaluation (i.e., validation and impact assessment) and updating of clinical prediction models.

METHODS

We systematically searched nine databases from January 2000 to January 2022 for articles in English with methodological recommendations for the post-derivation stages of interest. Qualitative analysis was used to summarize the 70 selected guidance papers.

RESULTS

Key aspects for validation are the assessment of statistical performance using measures for discrimination (e.g., C-statistic) and calibration (e.g., calibration-in-the-large and calibration slope). For assessing impact or usefulness in clinical decision-making, recent papers advise using decision-analytic measures (e.g., the Net Benefit) over simplistic classification measures that ignore clinical consequences (e.g., accuracy, overall Net Reclassification Index). Commonly recommended methods for model updating are recalibration (i.e., adjustment of intercept or baseline hazard and/or slope), revision (i.e., re-estimation of individual predictor effects), and extension (i.e., addition of new markers). Additional methodological guidance is needed for newer types of updating (e.g., meta-model and dynamic updating) and machine learning-based models.

CONCLUSION

Substantial guidance was found for model evaluation and more conventional updating of regression-based models. An important development in model evaluation is the introduction of a decision-analytic framework for assessing clinical usefulness. Consensus is emerging on methods for model updating.

Prognostic models for COVID-19 needed updating to warrant transportability over time and space

BACKGROUND

Supporting decisions for patients who present to the emergency department (ED) with COVID-19 requires accurate prognostication. We aimed to evaluate prognostic models for predicting outcomes in hospitalized patients with COVID-19, in different locations and across time.

METHODS

We included patients who presented to the ED with suspected COVID-19 and were admitted to 12 hospitals in the New York City (NYC) area and 4 large Dutch hospitals. We used second-wave patients who presented between September and December 2020 (2137 and 3252 in NYC and the Netherlands, respectively) to evaluate models that were developed on first-wave patients who presented between March and August 2020 (12,163 and 5831). We evaluated two prognostic models for in-hospital death: The Northwell COVID-19 Survival (NOCOS) model was developed on NYC data and the COVID Outcome Prediction in the Emergency Department (COPE) model was developed on Dutch data. These models were validated on subsequent second-wave data at the same site (temporal validation) and at the other site (geographic validation). We assessed model performance by the Area Under the receiver operating characteristic Curve (AUC), by the E-statistic, and by net benefit.

RESULTS

Twenty-eight-day mortality was considerably higher in the NYC first-wave data (21.0%), compared to the second-wave (10.1%) and the Dutch data (first wave 10.8%; second wave 10.0%). COPE discriminated well at temporal validation (AUC 0.82), with excellent calibration (E-statistic 0.8%). At geographic validation, discrimination was satisfactory (AUC 0.78), but with moderate over-prediction of mortality risk, particularly in higher-risk patients (E-statistic 2.9%). While discrimination was adequate when NOCOS was tested on second-wave NYC data (AUC 0.77), NOCOS systematically overestimated the mortality risk (E-statistic 5.1%). Discrimination in the Dutch data was good (AUC 0.81), but with over-prediction of risk, particularly in lower-risk patients (E-statistic 4.0%). Recalibration of COPE and NOCOS led to limited net benefit improvement in Dutch data, but to substantial net benefit improvement in NYC data.

CONCLUSIONS

NOCOS performed moderately worse than COPE, probably reflecting unique aspects of the early pandemic in NYC. Frequent updating of prognostic models is likely to be required for transportability over time and space during a dynamic pandemic.

Development and validation of self-monitoring auto-updating prognostic models of survival for hospitalized COVID-19 patients

Clinical prognostic models can assist patient care decisions. However, their performance can drift over time and location, necessitating model monitoring and updating. Despite rapid and significant changes during the pandemic, prognostic models for COVID-19 patients do not currently account for these drifts. We develop a framework for continuously monitoring and updating prognostic models and apply it to predict 28-day survival in COVID-19 patients. We use demographic, laboratory, and clinical data from electronic health records of 34912 hospitalized COVID-19 patients from March 2020 until May 2022 and compare three modeling methods. Model calibration performance drift is immediately detected with minor fluctuations in discrimination. The overall calibration on the prospective validation cohort is significantly improved when comparing the dynamically updated models against their static counterparts. Our findings suggest that, using this framework, models remain accurate and well-calibrated across various waves, variants, race and sex and yield positive net-benefits.