Performance of binary prediction models in high-correlation low-dimensional settings: a comparison of methods

Background Clinical prediction models are developed widely across medical disciplines. When predictors in such models are highly collinear, unexpected or spurious predictor-outcome associations may occur, thereby potentially reducing face-validity of the prediction model. Collinearity can be dealt with by exclusion of collinear predictors, but when there is no a priori motivation (besides collinearity) to include or exclude specific predictors, such an approach is arbitrary and possibly inappropriate. Methods We compare different methods to address collinearity, including shrinkage, dimensionality reduction, and constrained optimization. The effectiveness of these methods is illustrated via simulations. Results In the conducted simulations, no effect of collinearity was observed on predictive outcomes (AUC, R2, Intercept, Slope) across methods. However, a negative effect of collinearity on the stability of predictor selection was found, affecting all compared methods, but in particular methods that perform strong predictor selection (e.g., Lasso). Methods for which the included set of predictors remained most stable under increased collinearity were Ridge, PCLR, LAELR, and Dropout. Conclusions Based on the results, we would recommend refraining from data-driven predictor selection approaches in the presence of high collinearity, because of the increased instability of predictor selection, even in relatively high events-per-variable settings. The selection of certain predictors over others may disproportionally give the impression that included predictors have a stronger association with the outcome than excluded predictors.

Trends in clinical prediction models of stroke outcomes research outputs: a bibliometric analysis

Stroke is one of the leading causes of death and disability globally. Clinical models have been reported to predict stroke outcomes and thus, potentially guide clinical decisions. This study aimed to describe the global trends of research in clinical prediction models of stroke outcomes. A bibliometric analysis was conducted on clinical prediction models of stroke outcomes publications reported in Scopus from 2010 to 2019. Bibliographic data were extracted, quantitatively analysed, and visualized using VOSviewer software. A total of 6,364 publications were included in the final analysis. The number of published studies had steadily increased since 2010. “Stroke” and “Journal of Stroke and Cerebrovascular Diseases” were the journal with the most publications and citations. The most cited publications were by Lip et al. (2010) and Berkhemer et al. (2015). The United States of America (USA), China together with their institutions contributed most to the pool of publications in the field. Our study showed a steady increasing research activity in the clinical prediction of stroke outcomes since 2010, with saturation in recent years. The top articles were published in high-quality stroke-related journals and by high-income countries. There is a need to reinforce research capacities in the field by developing collaborative networks.

Facilitating safe discharge through predicting disease progression in moderate COVID-19: a prospective cohort study to develop and validate a clinical prediction model in resource-limited settings

ABSTRACTBackgroundIn locations where few people have received COVID-19 vaccines, health systems remain vulnerable to surges in SARS-CoV-2 infections. Tools to identify patients suitable for community-based management are urgently needed.MethodsWe prospectively recruited adults presenting to two hospitals in India with moderate symptoms of laboratory-confirmed COVID-19 in order to develop and validate a clinical prediction model to rule-out progression to supplemental oxygen requirement. The primary outcome was defined as any of the following: SpO2 < 94%; respiratory rate > 30 bpm; SpO2/FiO2 < 400; or death. We specified a priori that each model would contain three clinical parameters (age, sex and SpO2) and one of seven shortlisted biochemical biomarkers measurable using near-patient tests (CRP, D-dimer, IL-6, NLR, PCT, sTREM-1 or suPAR), to ensure the models would be suitable for resource-limited settings. We evaluated discrimination, calibration and clinical utility of the models in a temporal external validation cohort.Findings426 participants were recruited, of whom 89 (21·0%) met the primary outcome. 257 participants comprised the development cohort and 166 comprised the validation cohort. The three models containing NLR, suPAR or IL-6 demonstrated promising discrimination (c-statistics: 0·72 to 0·74) and calibration (calibration slopes: 1·01 to 1·05) in the validation cohort, and provided greater utility than a model containing the clinical parameters alone.InterpretationWe present three clinical prediction models that could help clinicians identify patients with moderate COVID-19 suitable for community-based management. The models are readily implementable and of particular relevance for locations with limited resources.FundingMédecins Sans Frontières, India.RESEARCH IN CONTEXTEvidence before this studyA living systematic review by Wynants et al. identified 137 COVID-19 prediction models, 47 of which were derived to predict whether patients with COVID-19 will have an adverse outcome. Most lacked external validation, relied on retrospective data, did not focus on patients with moderate disease, were at high risk of bias, and were not practical for use in resource-limited settings. To identify promising biochemical biomarkers which may have been evaluated independently of a prediction model and therefore not captured by this review, we searched PubMed on 1 June 2020 using synonyms of “SARS-CoV-2” AND [“biomarker” OR “prognosis”]. We identified 1,214 studies evaluating biochemical biomarkers of potential value in the prognostication of COVID-19 illness. In consultation with FIND (Geneva, Switzerland) we shortlisted seven candidates for evaluation in this study, all of which are measurable using near-patient tests which are either currently available or in late-stage development.Added value of this studyWe followed the TRIPOD guidelines to develop and validate three promising clinical prediction models to help clinicians identify which patients presenting with moderate COVID-19 can be safely managed in the community. Each model contains three easily ascertained clinical parameters (age, sex, and SpO2) and one biochemical biomarker (NLR, suPAR or IL-6), and would be practical for implementation in high-patient-throughput low resource settings. The models showed promising discrimination and calibration in the validation cohort. The inclusion of a biomarker test improved prognostication compared to a model containing the clinical parameters alone, and extended the range of contexts in which such a tool might provide utility to include situations when bed pressures are less critical, for example at earlier points in a COVID-19 surge.Implications of all the available evidencePrognostic models should be developed for clearly-defined clinical use-cases. We report the development and temporal validation of three clinical prediction models to rule-out progression to supplemental oxygen requirement amongst patients presenting with moderate COVID-19. The models are readily implementable and should prove useful in triage and resource allocation. We provide our full models to enable independent validation.

Machine learning-based clinical outcome prediction in surgery for acromegaly

Purpose Biochemical remission (BR), gross total resection (GTR), and intraoperative cerebrospinal fluid (CSF) leaks are important metrics in transsphenoidal surgery for acromegaly, and prediction of their likelihood using machine learning would be clinically advantageous. We aim to develop and externally validate clinical prediction models for outcomes after transsphenoidal surgery for acromegaly. Methods Using data from two registries, we develop and externally validate machine learning models for GTR, BR, and CSF leaks after endoscopic transsphenoidal surgery in acromegalic patients. For the model development a registry from Bologna, Italy was used. External validation was then performed using data from Zurich, Switzerland. Gender, age, prior surgery, as well as Hardy and Knosp classification were used as input features. Discrimination and calibration metrics were assessed. Results The derivation cohort consisted of 307 patients (43.3% male; mean [SD] age, 47.2 [12.7] years). GTR was achieved in 226 (73.6%) and BR in 245 (79.8%) patients. In the external validation cohort with 46 patients, 31 (75.6%) achieved GTR and 31 (77.5%) achieved BR. Area under the curve (AUC) at external validation was 0.75 (95% confidence interval: 0.59–0.88) for GTR, 0.63 (0.40–0.82) for BR, as well as 0.77 (0.62–0.91) for intraoperative CSF leaks. While prior surgery was the most important variable for prediction of GTR, age, and Hardy grading contributed most to the predictions of BR and CSF leaks, respectively. Conclusions Gross total resection, biochemical remission, and CSF leaks remain hard to predict, but machine learning offers potential in helping to tailor surgical therapy. We demonstrate the feasibility of developing and externally validating clinical prediction models for these outcomes after surgery for acromegaly and lay the groundwork for development of a multicenter model with more robust generalization.

Developing clinical prediction models when adhering to minimum sample size recommendations: The importance of quantifying bootstrap variability in tuning parameters and predictive performance

Recent minimum sample size formula (Riley et al.) for developing clinical prediction models help ensure that development datasets are of sufficient size to minimise overfitting. While these criteria are known to avoid excessive overfitting on average, the extent of variability in overfitting at recommended sample sizes is unknown. We investigated this through a simulation study and empirical example to develop logistic regression clinical prediction models using unpenalised maximum likelihood estimation, and various post-estimation shrinkage or penalisation methods. While the mean calibration slope was close to the ideal value of one for all methods, penalisation further reduced the level of overfitting, on average, compared to unpenalised methods. This came at the cost of higher variability in predictive performance for penalisation methods in external data. We recommend that penalisation methods are used in data that meet, or surpass, minimum sample size requirements to further mitigate overfitting, and that the variability in predictive performance and any tuning parameters should always be examined as part of the model development process, since this provides additional information over average (optimism-adjusted) performance alone. Lower variability would give reassurance that the developed clinical prediction model will perform well in new individuals from the same population as was used for model development.