Minimum sample size for external validation of a clinical prediction model with a binary outcome

Modern Learning from Big Data in Critical Care: Primum Non Nocere

Large and complex data sets are increasingly available for research in critical care. To analyze these data, researchers use techniques commonly referred to as statistical learning or machine learning (ML). The latter is known for large successes in the field of diagnostics, for example, by identification of radiological anomalies. In other research areas, such as clustering and prediction studies, there is more discussion regarding the benefit and efficiency of ML techniques compared with statistical learning. In this viewpoint, we aim to explain commonly used statistical learning and ML techniques and provide guidance for responsible use in the case of clustering and prediction questions in critical care. Clustering studies have been increasingly popular in critical care research, aiming to inform how patients can be characterized, classified, or treated differently. An important challenge for clustering studies is to ensure and assess generalizability. This limits the application of findings in these studies toward individual patients. In the case of predictive questions, there is much discussion as to what algorithm should be used to most accurately predict outcome. Aspects that determine usefulness of ML, compared with statistical techniques, include the volume of the data, the dimensionality of the preferred model, and the extent of missing data. There are areas in which modern ML methods may be preferred. However, efforts should be made to implement statistical frameworks (e.g., for dealing with missing data or measurement error, both omnipresent in clinical data) in ML methods. To conclude, there are important opportunities but also pitfalls to consider when performing clustering or predictive studies with ML techniques. We advocate careful valuation of new data-driven findings. More interaction is needed between the engineer mindset of experts in ML methods, the insight in bias of epidemiologists, and the probabilistic thinking of statisticians to extract as much information and knowledge from data as possible, while avoiding harm.

Discrimination and calibration of a prediction model for mortality is decreased in secondary transferred patients: a validation in the TraumaRegister DGU

INTRODUCTION

The Revised Injury Severity Classification II (RISC II) score represents a data-derived score that aims to predict mortality in severely injured patients. The aim of this study was to assess the discrimination and calibration of RISC II in secondary transferred polytrauma patients.

METHODS

This study was performed on the multicentre database of the TraumaRegister DGU. Inclusion criteria included Injury Severity Score (ISS)≥9 points and complete demographic data. Exclusion criteria included patients with 'do not resuscitate' orders or late transfers (>24 hours after initial trauma). Patients were stratified based on way of admission into patients transferred to a European trauma centre after initial treatment in another hospital (group Tr) and primary admitted patients who were not transferred out (group P). The RISC II score was calculated within each group at admission after secondary transfer (group Tr) and at primary admission (group P) and compared with the observed mortality rate. The calibration and discrimination of prediction were analysed.

RESULTS

Group P included 116 112 (91%) patients and group Tr included 11 604 (9%) patients. The study population was predominantly male (n=86 280, 70.1%), had a mean age of 53.2 years and a mean ISS of 20.7 points. Patients in group Tr were marginally older (54 years vs 52 years) and a had slightly higher ISS (21.5 points vs 20.1 points). Median time from accident site to hospital admission was 60 min in group P and 241 min (4 hours) in group Tr. Observed and predicted mortality based on RISC II were nearly identical in group P (10.9% and 11.0%, respectively) but predicted mortality was worse (13.4%) than observed mortality (11.1%) in group Tr.

CONCLUSION

The way of admission alters the calibration of prediction models for mortality in polytrauma patients. Mortality prediction in secondary transferred polytrauma patients should be calculated separately from primary admitted polytrauma patients.

Minimum sample size calculations for external validation of a clinical prediction model with a time-to-event outcome

Previous articles in Statistics in Medicine describe how to calculate the sample size required for external validation of prediction models with continuous and binary outcomes. The minimum sample size criteria aim to ensure precise estimation of key measures of a model's predictive performance, including measures of calibration, discrimination, and net benefit. Here, we extend the sample size guidance to prediction models with a time-to-event (survival) outcome, to cover external validation in datasets containing censoring. A simulation-based framework is proposed, which calculates the sample size required to target a particular confidence interval width for the calibration slope measuring the agreement between predicted risks (from the model) and observed risks (derived using pseudo-observations to account for censoring) on the log cumulative hazard scale. Precise estimation of calibration curves, discrimination, and net-benefit can also be checked in this framework. The process requires assumptions about the validation population in terms of the (i) distribution of the model's linear predictor and (ii) event and censoring distributions. Existing information can inform this; in particular, the linear predictor distribution can be approximated using the C-index or Royston's D statistic from the model development article, together with the overall event risk. We demonstrate how the approach can be used to calculate the sample size required to validate a prediction model for recurrent venous thromboembolism. Ideally the sample size should ensure precise calibration across the entire range of predicted risks, but must at least ensure adequate precision in regions important for clinical decision-making. Stata and R code are provided.

Development of a clinical prediction model to inform clinical decision making for classification of patients with sciatica, based on their clinical characteristics, in the Greek health system: protocol for a prospective predictive exploratory study

INTRODUCTION

Sciatica is one of the most common reasons for seeking healthcare for musculoskeletal pain. Sciatica is primarily considered as neuropathic in nature when neural tissue in the low back is compromised, but sometimes other non-neural structures may be involved. Appropriate assessment and management are important for patients with sciatica. Therapists use several outcome measures to assess patients to inform selection of the most suitable treatment. There is limited evidence for the best treatment of sciatica, and this is likely contributed to by having no reliable algorithm to categorise patients based on their clinical characteristics to inform physiotherapy treatment. The purpose of this study is to develop a clinical prediction model to categorise patients with sciatica, in terms of early clinical outcome, based on their initial clinical characteristics.

METHODS AND ANALYSIS

A prospective observational multicentre design will recruit consecutive patients (n=467) with sciatica referred for physiotherapy. Each patient will be evaluated to determine whether or not they will be accepted into the study by answering some questions that will confirm the study's eligibility criteria. Patients' basic characteristics, patient-reported outcome measures and performance-based measures will be collected at baseline from multiple sites in the Greek territory using this same protocol, prior to commencement of treatment. The main researcher of this study will be responsible for data collection in all sites. On completion of the standard referred physiotherapy treatment after 3 weeks' time, participants will be asked by telephone to evaluate their outcome using the Global Perceived Effect Scale. For the descriptive statistical analysis, the continuous variables will be expressed in the form of 'mean' and 'SD'. In order to assess the prognostic value of each predictor, in terms of the level of improvement or worsening of the symptoms, multiple variable regression analysis will be used.

ETHICS AND DISSEMINATION

Τhis study is approved from the Ethics and Deontology Committee of the University of West Attica, Athens, Greece, protocol number: 38313-09/06/2020, 10226-10/02/2021. The study's findings will be published in a peer-reviewed journal and disseminated at national and international conferences and through social media.

PROSPERO REGISTRATION NUMBER

CRD42020168467.

Guidelines and quality criteria for artificial intelligence-based prediction models in healthcare: a scoping review

While the opportunities of ML and AI in healthcare are promising, the growth of complex data-driven prediction models requires careful quality and applicability assessment before they are applied and disseminated in daily practice. This scoping review aimed to identify actionable guidance for those closely involved in AI-based prediction model (AIPM) development, evaluation and implementation including software engineers, data scientists, and healthcare professionals and to identify potential gaps in this guidance. We performed a scoping review of the relevant literature providing guidance or quality criteria regarding the development, evaluation, and implementation of AIPMs using a comprehensive multi-stage screening strategy. PubMed, Web of Science, and the ACM Digital Library were searched, and AI experts were consulted. Topics were extracted from the identified literature and summarized across the six phases at the core of this review: (1) data preparation, (2) AIPM development, (3) AIPM validation, (4) software development, (5) AIPM impact assessment, and (6) AIPM implementation into daily healthcare practice. From 2683 unique hits, 72 relevant guidance documents were identified. Substantial guidance was found for data preparation, AIPM development and AIPM validation (phases 1-3), while later phases clearly have received less attention (software development, impact assessment and implementation) in the scientific literature. The six phases of the AIPM development, evaluation and implementation cycle provide a framework for responsible introduction of AI-based prediction models in healthcare. Additional domain and technology specific research may be necessary and more practical experience with implementing AIPMs is needed to support further guidance.

Recommendations on compiling test datasets for evaluating artificial intelligence solutions in pathology

Artificial intelligence (AI) solutions that automatically extract information from digital histology images have shown great promise for improving pathological diagnosis. Prior to routine use, it is important to evaluate their predictive performance and obtain regulatory approval. This assessment requires appropriate test datasets. However, compiling such datasets is challenging and specific recommendations are missing. A committee of various stakeholders, including commercial AI developers, pathologists, and researchers, discussed key aspects and conducted extensive literature reviews on test datasets in pathology. Here, we summarize the results and derive general recommendations on compiling test datasets. We address several questions: Which and how many images are needed? How to deal with low-prevalence subsets? How can potential bias be detected? How should datasets be reported? What are the regulatory requirements in different countries? The recommendations are intended to help AI developers demonstrate the utility of their products and to help pathologists and regulatory agencies verify reported performance measures. Further research is needed to formulate criteria for sufficiently representative test datasets so that AI solutions can operate with less user intervention and better support diagnostic workflows in the future.

External validation of Machine Learning models for COVID-19 detection based on Complete Blood Count

PURPOSE

The rRT-PCR for COVID-19 diagnosis is affected by long turnaround time, potential shortage of reagents, high false-negative rates and high costs. Routine hematochemical tests are a faster and less expensive alternative for diagnosis. Thus, Machine Learning (ML) has been applied to hematological parameters to develop diagnostic tools and help clinicians in promptly managing positive patients. However, few ML models have been externally validated, making their real-world applicability unclear.

METHODS

We externally validate 6 state-of-the-art diagnostic ML models, based on Complete Blood Count (CBC) and trained on a dataset encompassing 816 COVID-19 positive cases. The external validation was performed based on two datasets, collected at two different hospitals in northern Italy and encompassing 163 and 104 COVID-19 positive cases, in terms of both error rate and calibration.

RESULTS AND CONCLUSION

We report an average AUC of 95% and average Brier score of 0.11, out-performing existing ML methods, and showing good cross-site transportability. The best performing model (SVM) reported an average AUC of 97.5% (Sensitivity: 87.5%, Specificity: 94%), comparable with the performance of RT-PCR, and was also the best calibrated. The validated models can be useful in the early identification of potential COVID-19 patients, due to the rapid availability of CBC exams, and in multiple test settings.

The Safety INdEx of Prehospital On Scene Triage (SINEPOST) study: the development and validation of a risk prediction model to support ambulance clinical transport decisions on-scene-a protocol

BACKGROUND

Demand for both the ambulance service and the emergency department (ED) is rising every year and when this demand is excessive in both systems, ambulance crews queue at the ED waiting to hand patients over. Some transported ambulance patients are 'low-acuity' and do not require the treatment of the ED. However, paramedics can find it challenging to identify these patients accurately. Decision support tools have been developed using expert opinion to help identify these low acuity patients but have failed to show a benefit beyond regular decision-making. Predictive algorithms may be able to build accurate models, which can be used in the field to support the decision not to take a low-acuity patient to an ED.

METHODS AND ANALYSIS

All patients in Yorkshire who were transported to the ED by ambulance between July 2019 and February 2020 will be included. Ambulance electronic patient care record (ePCR) clinical data will be used as candidate predictors for the model. These will then be linked to the corresponding ED record, which holds the outcome of a 'non-urgent attendance'. The estimated sample size is 52,958, with 4767 events and an EPP of 7.48. An XGBoost algorithm will be used for model development. Initially, a model will be derived using all the data and the apparent performance will be assessed. Then internal-external validation will use non-random nested cross-validation (CV) with test sets held out for each ED (spatial validation). After all models are created, a random-effects meta-analysis will be undertaken. This will pool performance measures such as goodness of fit, discrimination and calibration. It will also generate a prediction interval and measure heterogeneity between clusters. The performance of the full model will be updated with the pooled results.

DISCUSSION

Creating a risk prediction model in this area will lead to further development of a clinical decision support tool that ensures every ambulance patient can get to the right place of care, first time. If this study is successful, it could help paramedics evaluate the benefit of transporting a patient to the ED before they leave the scene. It could also reduce congestion in the urgent and emergency care system.

TRIAL REGISTRATION

This study was retrospectively registered with the ISRCTN: 12121281.

Evaluating live performance of machine learning based prediction models for different clinical risks: a study of live systems in different hospitals (Preprint)

Background

Machine learning algorithms are currently used in a wide array of clinical domains to produce models that can predict clinical risk events. Most models are developed and evaluated with retrospective data, very few are evaluated in a clinical workflow, and even fewer report performances in different hospitals. In this study, we provide detailed evaluations of clinical risk prediction models in live clinical workflows for three different use cases in three different hospitals.

Objective

The main objective of this study was to evaluate clinical risk prediction models in live clinical workflows and compare their performance in these setting with their performance when using retrospective data. We also aimed at generalizing the results by applying our investigation to three different use cases in three different hospitals.

Methods

We trained clinical risk prediction models for three use cases (ie, delirium, sepsis, and acute kidney injury) in three different hospitals with retrospective data. We used machine learning and, specifically, deep learning to train models that were based on the Transformer model. The models were trained using a calibration tool that is common for all hospitals and use cases. The models had a common design but were calibrated using each hospital’s specific data. The models were deployed in these three hospitals and used in daily clinical practice. The predictions made by these models were logged and correlated with the diagnosis at discharge. We compared their performance with evaluations on retrospective data and conducted cross-hospital evaluations.

Results

The performance of the prediction models with data from live clinical workflows was similar to the performance with retrospective data. The average value of the area under the receiver operating characteristic curve (AUROC) decreased slightly by 0.6 percentage points (from 94.8% to 94.2% at discharge). The cross-hospital evaluations exhibited severely reduced performance: the average AUROC decreased by 8 percentage points (from 94.2% to 86.3% at discharge), which indicates the importance of model calibration with data from the deployment hospital.

Conclusions

Calibrating the prediction model with data from different deployment hospitals led to good performance in live settings. The performance degradation in the cross-hospital evaluation identified limitations in developing a generic model for different hospitals. Designing a generic process for model development to generate specialized prediction models for each hospital guarantees model performance in different hospitals.

Risk of caesarean delivery in labour induction: a systematic review and external validation of predictive models

BACKGROUND

Despite the existence of numerous published models predicting the risk of caesarean delivery in women undergoing induction of labour (IOL), validated models are scarce.

OBJECTIVES

To systematically review and externally assess the predictive capacity of caesarean delivery risk models in women undergoing IOL.

SEARCH STRATEGY

Studies published up to 15 January 2021 were identified through PubMed, CINAHL, Scopus and ClinicalTrials.gov, without temporal or language restrictions.

SELECTION CRITERIA

Studies describing the derivation of new models for predicting the risk of caesarean delivery in labour induction.

DATA COLLECTION AND ANALYSIS

Three authors independently screened the articles and assessed the risk of bias (ROB) according to the prediction model risk of bias assessment tool (PROBAST). External validation was performed in a prospective cohort of 468 pregnancies undergoing IOL from February 2019 to August 2020. The predictive capacity of the models was assessed by creating areas under the receiver operating characteristic curve (AUCs), calibration plots and decision curve analysis (DCA).

MAIN RESULTS

Fifteen studies met the eligibility criteria; 12 predictive models were validated. The quality of most of the included studies was not adequate. The AUC of the models varied from 0.520 to 0.773. The three models with the best discriminative capacity were those of Levine et al. (AUC 0.773, 95% CI 0.720-0.827), Hernández et al. (AUC 0.762, 95% CI 0.715-0.809) and Rossi et al. (AUC 0.752, 95% CI 0.707-0.797).

CONCLUSIONS

Predictive capacity and methodological quality were limited; therefore, we cannot currently recommend the use of any of the models for decision making in clinical practice.

TWEETABLE ABSTRACT

Predictive models that predict the risk of cesarean section in labor inductions are currently not applicable.

The importance of being external. methodological insights for the external validation of machine learning models in medicine

Background and Objective Medical machine learning (ML) models tend to perform better on data from the same cohort than on new data, often due to overfitting, or co-variate shifts. For these reasons, external validation (EV) is a necessary practice in the evaluation of medical ML. However, there is still a gap in the literature on how to interpret EV results and hence assess the robustness of ML models.

METHODS

We fill this gap by proposing a meta-validation method, to assess the soundness of EV procedures. In doing so, we complement the usual way to assess EV by considering both dataset cardinality, and the similarity of the EV dataset with respect to the training set. We then investigate how the notions of cardinality and similarity can be used to inform on the reliability of a validation procedure, by integrating them into two summative data visualizations.

RESULTS

We illustrate our methodology by applying it to the validation of a state-of-the-art COVID-19 diagnostic model on 8 EV sets, collected across 3 different continents. The model performance was moderately impacted by data similarity (Pearson ρ = 0.38, p< 0.001). In the EV, the validated model reported good AUC (average: 0.84), acceptable calibration (average: 0.17) and utility (average: 0.50). The validation datasets were adequate in terms of dataset cardinality and similarity, thus suggesting the soundness of the results. We also provide a qualitative guideline to evaluate the reliability of validation procedures, and we discuss the importance of proper external validation in light of the obtained results.

CONCLUSIONS

In this paper, we propose a novel, lean methodology to: 1) study how the similarity between training and validation sets impacts the generalizability of a ML model; 2) assess the soundness of EV evaluations along three complementary performance dimensions: discrimination, utility and calibration; 3) draw conclusions on the robustness of the model under validation. We applied this methodology to a state-of-the-art model for the diagnosis of COVID-19 from routine blood tests, and showed how to interpret the results in light of the presented framework.

Prediction models for diagnosis and prognosis of covid-19: systematic review and critical appraisal

OBJECTIVE

To review and appraise the validity and usefulness of published and preprint reports of prediction models for diagnosing coronavirus disease 2019 (covid-19) in patients with suspected infection, for prognosis of patients with covid-19, and for detecting people in the general population at increased risk of covid-19 infection or being admitted to hospital with the disease.

DESIGN

Living systematic review and critical appraisal by the COVID-PRECISE (Precise Risk Estimation to optimise covid-19 Care for Infected or Suspected patients in diverse sEttings) group.

DATA SOURCES

PubMed and Embase through Ovid, up to 1 July 2020, supplemented with arXiv, medRxiv, and bioRxiv up to 5 May 2020.

STUDY SELECTION

Studies that developed or validated a multivariable covid-19 related prediction model.

DATA EXTRACTION

At least two authors independently extracted data using the CHARMS (critical appraisal and data extraction for systematic reviews of prediction modelling studies) checklist; risk of bias was assessed using PROBAST (prediction model risk of bias assessment tool).

RESULTS

37 421 titles were screened, and 169 studies describing 232 prediction models were included. The review identified seven models for identifying people at risk in the general population; 118 diagnostic models for detecting covid-19 (75 were based on medical imaging, 10 to diagnose disease severity); and 107 prognostic models for predicting mortality risk, progression to severe disease, intensive care unit admission, ventilation, intubation, or length of hospital stay. The most frequent types of predictors included in the covid-19 prediction models are vital signs, age, comorbidities, and image features. Flu-like symptoms are frequently predictive in diagnostic models, while sex, C reactive protein, and lymphocyte counts are frequent prognostic factors. Reported C index estimates from the strongest form of validation available per model ranged from 0.71 to 0.99 in prediction models for the general population, from 0.65 to more than 0.99 in diagnostic models, and from 0.54 to 0.99 in prognostic models. All models were rated at high or unclear risk of bias, mostly because of non-representative selection of control patients, exclusion of patients who had not experienced the event of interest by the end of the study, high risk of model overfitting, and unclear reporting. Many models did not include a description of the target population (n=27, 12%) or care setting (n=75, 32%), and only 11 (5%) were externally validated by a calibration plot. The Jehi diagnostic model and the 4C mortality score were identified as promising models.

CONCLUSION

Prediction models for covid-19 are quickly entering the academic literature to support medical decision making at a time when they are urgently needed. This review indicates that almost all pubished prediction models are poorly reported, and at high risk of bias such that their reported predictive performance is probably optimistic. However, we have identified two (one diagnostic and one prognostic) promising models that should soon be validated in multiple cohorts, preferably through collaborative efforts and data sharing to also allow an investigation of the stability and heterogeneity in their performance across populations and settings. Details on all reviewed models are publicly available at https://www.covprecise.org/. Methodological guidance as provided in this paper should be followed because unreliable predictions could cause more harm than benefit in guiding clinical decisions. Finally, prediction model authors should adhere to the TRIPOD (transparent reporting of a multivariable prediction model for individual prognosis or diagnosis) reporting guideline.

SYSTEMATIC REVIEW REGISTRATION

Protocol https://osf.io/ehc47/, registration https://osf.io/wy245.

READERS' NOTE

This article is a living systematic review that will be updated to reflect emerging evidence. Updates may occur for up to two years from the date of original publication. This version is update 3 of the original article published on 7 April 2020 (BMJ 2020;369:m1328). Previous updates can be found as data supplements (https://www.bmj.com/content/369/bmj.m1328/related#datasupp). When citing this paper please consider adding the update number and date of access for clarity.