Show Your Work: Responsible Model Reporting in Health Care Artificial Intelligence.

Standardized and thorough model reporting is an integral component in the development and deployment of machine learning models in health care. Model reporting includes sharing multiple model performance metrics and incorporating metadata to provide the necessary context for model evaluation. Thorough model reporting addresses common concerns about artificial intelligence in health care including model explainability, transparency, fairness, and generalizability. All stages in the model development lifecycle, from initial design to data capture to model deployment, can be communicated openly to stakeholders with responsible model reporting. Physician involvement throughout these processes can ensure clinical concerns and potential consequences are considered.

Considerations in the reliability and fairness audits of predictive models for advance care planning.

Multiple reporting guidelines for artificial intelligence (AI) models in healthcare recommend that models be audited for reliability and fairness. However, there is a gap of operational guidance for performing reliability and fairness audits in practice. Following guideline recommendations, we conducted a reliability audit of two models based on model performance and calibration as well as a fairness audit based on summary statistics, subgroup performance and subgroup calibration. We assessed the Epic End-of-Life (EOL) Index model and an internally developed Stanford Hospital Medicine (HM) Advance Care Planning (ACP) model in 3 practice settings: Primary Care, Inpatient Oncology and Hospital Medicine, using clinicians' answers to the surprise question ("Would you be surprised if [patient X] passed away in [Y years]?") as a surrogate outcome. For performance, the models had positive predictive value (PPV) at or above 0.76 in all settings. In Hospital Medicine and Inpatient Oncology, the Stanford HM ACP model had higher sensitivity (0.69, 0.89 respectively) than the EOL model (0.20, 0.27), and better calibration (O/E 1.5, 1.7) than the EOL model (O/E 2.5, 3.0). The Epic EOL model flagged fewer patients (11%, 21% respectively) than the Stanford HM ACP model (38%, 75%). There were no differences in performance and calibration by sex. Both models had lower sensitivity in Hispanic/Latino male patients with Race listed as "Other." 10 clinicians were surveyed after a presentation summarizing the audit. 10/10 reported that summary statistics, overall performance, and subgroup performance would affect their decision to use the model to guide care; 9/10 said the same for overall and subgroup calibration. The most commonly identified barriers for routinely conducting such reliability and fairness audits were poor demographic data quality and lack of data access. This audit required 115 person-hours across 8-10 months. Our recommendations for performing reliability and fairness audits include verifying data validity, analyzing model performance on intersectional subgroups, and collecting clinician-patient linkages as necessary for label generation by clinicians. Those responsible for AI models should require such audits before model deployment and mediate between model auditors and impacted stakeholders.

Predicting PBT and CMR properties of substances of very high concern (SVHCs) using QSAR models, and application for K-REACH.

Quantitative structure-activity relationship (QSAR) models have been applied to predict a variety of toxicity endpoints. Their performance needs to be validated, in a variety of cases, to increase their applicability to chemical regulation. Using the data set of substances of very high concern (SVHCs), the performance of QSAR models were evaluated to predict the persistence and bioaccumulation of PBT, and the carcinogenicity and mutagenicity of CMR. BIOWIN and Toxtree showed higher sensitivity than other QSAR models - the former for persistence and bioaccumulation, the latter for carcinogenicity. In terms of mutagenicity, the sensitivities of QSAR models were underestimated, Toxtree was more accurate and specific than lazy structure-activity relationships (LAZARs) and Computer Assisted Evaluation of industrial chemical Substances According to Regulations (CAESAR). Using the weight of evidence (WoE) approach, which integrates results of individual QSAR models, enhanced the sensitivity of each toxicity endpoint. On the basis of obtained results, in particular the prediction of persistence and bioaccumulation by KOWWIN, a conservative criterion is recommended of log Kow greater than 4.5 in K-REACH, without an upper limit. This study suggests that reliable production of toxicity data by QSAR models is facilitated by a better understanding of the performance of these models.

Methodological standards for the development and evaluation of clinical prediction rules: a review of the literature.

Clinical prediction rules (CPRs) that predict the absolute risk of a clinical condition or future outcome for individual patients are abundant in the medical literature; however, systematic reviews have demonstrated shortcomings in the methodological quality and reporting of prediction studies. To maximise the potential and clinical usefulness of CPRs, they must be rigorously developed and validated, and their impact on clinical practice and patient outcomes must be evaluated. This review aims to present a comprehensive overview of the stages involved in the development, validation and evaluation of CPRs, and to describe in detail the methodological standards required at each stage, illustrated with examples where appropriate. Important features of the study design, statistical analysis, modelling strategy, data collection, performance assessment, CPR presentation and reporting are discussed, in addition to other, often overlooked aspects such as the acceptability, cost-effectiveness and longer-term implementation of CPRs, and their comparison with clinical judgement. Although the development and evaluation of a robust, clinically useful CPR is anything but straightforward, adherence to the plethora of methodological standards, recommendations and frameworks at each stage will assist in the development of a rigorous CPR that has the potential to contribute usefully to clinical practice and decision-making and have a positive impact on patient care.

Computational models for the assessment of manufactured nanomaterials: Development of model reporting standards and mapping of the model landscape.

Different types of computational models have been developed for predicting the biokinetics, environmental fate, exposure levels and toxicological effects of chemicals and manufactured nanomaterials (MNs). However, these models are not described in a consistent manner in the scientific literature, which is one of the barriers to their broader use and acceptance, especially for regulatory purposes. Quantitative structure-activity relationships (QSARs) are in silico models based on the assumption that the activity of a substance is related to its chemical structure. These models can be used to provide information on (eco)toxicological effects in hazard assessment. In an environmental risk assessment, environmental exposure models can be used to estimate the predicted environmental concentration (PEC). In addition, physiologically based kinetic (PBK) models can be used in various ways to support a human health risk assessment. In this paper, we first propose model reporting templates for systematically and transparently describing models that could potentially be used to support regulatory risk assessments of MNs, for example under the REACH regulation. The model reporting templates include (a) the adaptation of the QSAR Model Reporting Format (QMRF) to report models for MNs, and (b) the development of a model reporting template for PBK and environmental exposure models applicable to MNs. Second, we show the usefulness of these templates to report different models, resulting in an overview of the landscape of available computational models for MNs.

Key steps and common pitfalls in developing and validating risk models.

Models for estimating an individual's risk of having or developing a disease are abundant in the medical literature, yet many do not meet the methodological standards that have been set to maximise generalisability and utility. This paper presents an overview of ten steps from the conception of the study to the implementation of the risk model and discusses common pitfalls. We discuss crucial aspects of study design, data collection, model development and performance evaluation, and discuss how to bring the model to clinical practice. TWEETABLE ABSTRACT: We present an overview of ten key steps for the development of risk models and discuss common pitfalls.