Comparison of dynamic updating strategies for clinical prediction models.
Ontology highlight
ABSTRACT: Background
Prediction models inform many medical decisions, but their performance often deteriorates over time. Several discrete-time update strategies have been proposed in the literature, including model recalibration and revision. However, these strategies have not been compared in the dynamic updating setting.Methods
We used post-lung transplant survival data during 2010-2015 and compared the Brier Score (BS), discrimination, and calibration of the following update strategies: (1) never update, (2) update using the closed testing procedure proposed in the literature, (3) always recalibrate the intercept, (4) always recalibrate the intercept and slope, and (5) always refit/revise the model. In each case, we explored update intervals of every 1, 2, 4, and 8 quarters. We also examined how the performance of the update strategies changed as the amount of old data included in the update (i.e., sliding window length) increased.Results
All methods of updating the model led to meaningful improvement in BS relative to never updating. More frequent updating yielded better BS, discrimination, and calibration, regardless of update strategy. Recalibration strategies led to more consistent improvements and less variability over time compared to the other updating strategies. Using longer sliding windows did not substantially impact the recalibration strategies, but did improve the discrimination and calibration of the closed testing procedure and model revision strategies.Conclusions
Model updating leads to improved BS, with more frequent updating performing better than less frequent updating. Model recalibration strategies appeared to be the least sensitive to the update interval and sliding window length.
SUBMITTER: Schnellinger EM
PROVIDER: S-EPMC8647501 | biostudies-literature |
REPOSITORIES: biostudies-literature
Similar Datasets
Project description:There is growing interest in developing clinical prediction models (CPMs) to aid local healthcare decision-making. Frequently, these CPMs are developed in isolation across different populations, with repetitive de novo derivation a common modelling strategy. However, this fails to utilise all available information and does not respond to changes in health processes through time and space. Alternatively, model updating techniques have previously been proposed that adjust an existing CPM to suit the new population, but these techniques are restricted to a single model. Therefore, we aimed to develop a generalised method for updating and aggregating multiple CPMs. The proposed "hybrid method" re-calibrates multiple CPMs using stacked regression while concurrently revising specific covariates using individual participant data (IPD) under a penalised likelihood. The performance of the hybrid method was compared with existing methods in a clinical example of mortality risk prediction after transcatheter aortic valve implantation, and in 2 simulation studies. The simulation studies explored the effect of sample size and between-population-heterogeneity on the method, with each representing a situation of having multiple distinct CPMs and 1 set of IPD. When the sample size of the IPD was small, stacked regression and the hybrid method had comparable but highest performance across modelling methods. Conversely, in large IPD samples, development of a new model and the hybrid method gave the highest performance. Hence, the proposed strategy can inform the choice between utilising existing CPMs or developing a model de novo, thereby incorporating IPD, existing research, and prior (clinical) knowledge into the modelling strategy.
| S-EPMC5873448 | biostudies-literature
Project description:Model calibration, critical to the success and safety of clinical prediction models, deteriorates over time in response to the dynamic nature of clinical environments. To support informed, data-driven model updating strategies, we present and evaluate a calibration drift detection system. Methods are developed for maintaining dynamic calibration curves with optimized online stochastic gradient descent and for detecting increasing miscalibration with adaptive sliding windows. These methods are generalizable to support diverse prediction models developed using a variety of learning algorithms and customizable to address the unique needs of clinical use cases. In both simulation and case studies, our system accurately detected calibration drift. When drift is detected, our system further provides actionable alerts by including information on a window of recent data that may be appropriate for model updating. Simulations showed these windows were primarily composed of data accruing after drift onset, supporting the potential utility of the windows for model updating. By promoting model updating as calibration deteriorates rather than on pre-determined schedules, implementations of our drift detection system may minimize interim periods of insufficient model accuracy and focus analytic resources on those models most in need of attention.
| S-EPMC8627243 | biostudies-literature
Project description:BackgroundClinical 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.MethodsWe 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.ResultsKey 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.ConclusionSubstantial 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.
| S-EPMC9742671 | biostudies-literature
Project description:BACKGROUND: Scientific knowledge is in constant change. The flow of new information requires a frequent re-evaluation of the available research results. Clinical practice guidelines (CPGs) are not exempted from this phenomenon and need to be kept updated to maintain the validity of their recommendations. The objective of our review is to systematically identify, describe and assess strategies for monitoring and updating CPGs. STUDY DESIGN AND SETTING: We conducted a systematic review of studies evaluating one or more methods of updating (with or without monitoring) CPGs or recommendations. We searched MEDLINE (PubMed) and The Cochrane Methodology Register (The Cochrane Library) from 1966 to June 2012. Additionally, we hand-searched reference lists of the included studies and the Guidelines International Network book of abstracts. If necessary, we contacted study authors to obtain additional information. RESULTS: We included a total of eight studies. Four evaluated if CPGs were out of date, three updated CPGs, and one continuously monitored and updated CPGs. The most detailed reported phase of the process was the identification of new evidence. As opposed to studies updating guidelines, studies evaluating if CPGs were out of date applied restricted searches. Only one study compared a restricted versus an exhaustive search suggesting that a restricted search is sufficient to assess recommendations' Validity. One study analyzed the survival time of CPGs and suggested that these should be reassessed every three years. CONCLUSIONS: There is limited evidence about the optimal strategies for monitoring and updating clinical practice guidelines. A restricted search is likely to be sufficient to monitor new evidence and assess the need to update, however, more information is needed about the timing and type of search. Only the exhaustive search strategy has been assessed for the update of CPGs. The development and evaluation of more efficient strategies is needed to improve the timeliness and reduce the burden of maintaining the validity of CPGs.
| S-EPMC3520818 | biostudies-other
Project description:ObjectiveClinical prediction models require updating as performance deteriorates over time. We developed a testing procedure to select updating methods that minimizes overfitting, incorporates uncertainty associated with updating sample sizes, and is applicable to both parametric and nonparametric models.Materials and methodsWe describe a procedure to select an updating method for dichotomous outcome models by balancing simplicity against accuracy. We illustrate the test's properties on simulated scenarios of population shift and 2 models based on Department of Veterans Affairs inpatient admissions.ResultsIn simulations, the test generally recommended no update under no population shift, no update or modest recalibration under case mix shifts, intercept correction under changing outcome rates, and refitting under shifted predictor-outcome associations. The recommended updates provided superior or similar calibration to that achieved with more complex updating. In the case study, however, small update sets lead the test to recommend simpler updates than may have been ideal based on subsequent performance.DiscussionOur test's recommendations highlighted the benefits of simple updating as opposed to systematic refitting in response to performance drift. The complexity of recommended updating methods reflected sample size and magnitude of performance drift, as anticipated. The case study highlights the conservative nature of our test.ConclusionsThis new test supports data-driven updating of models developed with both biostatistical and machine learning approaches, promoting the transportability and maintenance of a wide array of clinical prediction models and, in turn, a variety of applications relying on modern prediction tools.
| S-EPMC6857513 | biostudies-literature
Project description:Six recently published algorithms classify pneumonia patients presenting from the community into high- and low-risk groups for resistant bacteria. Our objective was to compare performance of these algorithms for identifying patients infected with bacteria resistant to traditional community-acquired pneumonia antibiotics.This was a retrospective study of consecutive adult patients diagnosed with pneumonia in an emergency department and subsequently hospitalized. Each patient was classified as high or low risk for resistant bacteria according to the following algorithms: original health care-associated pneumonia (HCAP) criteria, Summit criteria, Brito and Niederman strategy, Shorr model, Aliberti model, and Shindo model. The reference for comparison was detection of resistant bacteria, defined as methicillin-resistant Staphylococcus aureus or Gram-negative bacteria resistant to ceftriaxone or levofloxacin.A total of 614 patients were studied, including 36 (5.9%) with resistant bacteria. The HCAP criteria classified 304 (49.5%) patients as high risk, with an area under the receiver operating characteristic curve (AUC) of 0.63 (95% confidence interval [CI] = 0.54 to 0.72), sensitivity of 0.69 (95% CI = 0.52 to 0.83), and specificity of 0.52 (95% CI = 0.48 to 0.56). None of the other algorithms improved both sensitivity and specificity or significantly improved the AUC. Compared to the HCAP criteria, the Shorr and Aliberti models classified more patients as high risk, resulting in higher sensitivity and lower specificity. The Shindo model classified fewer patients as high risk, with lower sensitivity and higher specificity.All algorithms for identification of resistant bacteria included in this study had suboptimal performance to guide antibiotic selection. New strategies for selecting empirical antibiotics for community-onset pneumonia are necessary.
| S-EPMC4464985 | biostudies-literature
Project description:BackgroundGene expression profiling is being widely applied in cancer research to identify biomarkers for clinical endpoint prediction. Since RNA-seq provides a powerful tool for transcriptome-based applications beyond the limitations of microarrays, we sought to systematically evaluate the performance of RNA-seq-based and microarray-based classifiers in this MAQC-III/SEQC study for clinical endpoint prediction using neuroblastoma as a model.ResultsWe generate gene expression profiles from 498 primary neuroblastomas using both RNA-seq and 44 k microarrays. Characterization of the neuroblastoma transcriptome by RNA-seq reveals that more than 48,000 genes and 200,000 transcripts are being expressed in this malignancy. We also find that RNA-seq provides much more detailed information on specific transcript expression patterns in clinico-genetic neuroblastoma subgroups than microarrays. To systematically compare the power of RNA-seq and microarray-based models in predicting clinical endpoints, we divide the cohort randomly into training and validation sets and develop 360 predictive models on six clinical endpoints of varying predictability. Evaluation of factors potentially affecting model performances reveals that prediction accuracies are most strongly influenced by the nature of the clinical endpoint, whereas technological platforms (RNA-seq vs. microarrays), RNA-seq data analysis pipelines, and feature levels (gene vs. transcript vs. exon-junction level) do not significantly affect performances of the models.ConclusionsWe demonstrate that RNA-seq outperforms microarrays in determining the transcriptomic characteristics of cancer, while RNA-seq and microarray-based models perform similarly in clinical endpoint prediction. Our findings may be valuable to guide future studies on the development of gene expression-based predictive models and their implementation in clinical practice.
| S-EPMC4506430 | biostudies-literature
Project description:In visual search tasks, repeating features or the position of the target results in faster response times. Such inter-trial 'priming' effects occur not just for repetitions from the immediately preceding trial but also from trials further back. A paradigm known to produce particularly long-lasting inter-trial effects-of the target-defining feature, target position, and response (feature)-is the 'priming of pop-out' (PoP) paradigm, which typically uses sparse search displays and random swapping across trials of target- and distractor-defining features. However, the mechanisms underlying these inter-trial effects are still not well understood. To address this, we applied a modeling framework combining an evidence accumulation (EA) model with different computational updating rules of the model parameters (i.e., the drift rate and starting point of EA) for different aspects of stimulus history, to data from a (previously published) PoP study that had revealed significant inter-trial effects from several trials back for repetitions of the target color, the target position, and (response-critical) target feature. By performing a systematic model comparison, we aimed to determine which EA model parameter and which updating rule for that parameter best accounts for each inter-trial effect and the associated n-back temporal profile. We found that, in general, our modeling framework could accurately predict the n-back temporal profiles. Further, target color- and position-based inter-trial effects were best understood as arising from redistribution of a limited-capacity weight resource which determines the EA rate. In contrast, response-based inter-trial effects were best explained by a bias of the starting point towards the response associated with a previous target; this bias appeared largely tied to the position of the target. These findings elucidate how our cognitive system continually tracks, and updates an internal predictive model of, a number of separable stimulus and response parameters in order to optimize task performance.
| S-EPMC8445473 | biostudies-literature
Project description:ImportanceThe lack of standards in methods to reduce bias for clinical algorithms presents various challenges in providing reliable predictions and in addressing health disparities.ObjectiveTo evaluate approaches for reducing bias in machine learning models using a real-world clinical scenario.Design, setting, and participantsHealth data for this cohort study were obtained from the IBM MarketScan Medicaid Database. Eligibility criteria were as follows: (1) Female individuals aged 12 to 55 years with a live birth record identified by delivery-related codes from January 1, 2014, through December 31, 2018; (2) greater than 80% enrollment through pregnancy to 60 days post partum; and (3) evidence of coverage for depression screening and mental health services. Statistical analysis was performed in 2020.ExposuresBinarized race (Black individuals and White individuals).Main outcomes and measuresMachine learning models (logistic regression [LR], random forest, and extreme gradient boosting) were trained for 2 binary outcomes: postpartum depression (PPD) and postpartum mental health service utilization. Risk-adjusted generalized linear models were used for each outcome to assess potential disparity in the cohort associated with binarized race (Black or White). Methods for reducing bias, including reweighing, Prejudice Remover, and removing race from the models, were examined by analyzing changes in fairness metrics compared with the base models. Baseline characteristics of female individuals at the top-predicted risk decile were compared for systematic differences. Fairness metrics of disparate impact (DI, 1 indicates fairness) and equal opportunity difference (EOD, 0 indicates fairness).ResultsAmong 573 634 female individuals initially examined for this study, 314 903 were White (54.9%), 217 899 were Black (38.0%), and the mean (SD) age was 26.1 (5.5) years. The risk-adjusted odds ratio comparing White participants with Black participants was 2.06 (95% CI, 2.02-2.10) for clinically recognized PPD and 1.37 (95% CI, 1.33-1.40) for postpartum mental health service utilization. Taking the LR model for PPD prediction as an example, reweighing reduced bias as measured by improved DI and EOD metrics from 0.31 and -0.19 to 0.79 and 0.02, respectively. Removing race from the models had inferior performance for reducing bias compared with the other methods (PPD: DI = 0.61; EOD = -0.05; mental health service utilization: DI = 0.63; EOD = -0.04).Conclusions and relevanceClinical prediction models trained on potentially biased data may produce unfair outcomes on the basis of the chosen metrics. This study's results suggest that the performance varied depending on the model, outcome label, and method for reducing bias. This approach toward evaluating algorithmic bias can be used as an example for the growing number of researchers who wish to examine and address bias in their data and models.
| S-EPMC8050742 | biostudies-literature
Project description:Many previous studies on visual search have reported inter-trial effects, that is, observers respond faster when some target property, such as a defining feature or dimension, or the response associated with the target repeats versus changes across consecutive trial episodes. However, what processes drive these inter-trial effects is still controversial. Here, we investigated this question using a combination of Bayesian modeling of belief updating and evidence accumulation modeling in perceptual decision-making. In three visual singleton ('pop-out') search experiments, we explored how the probability of the response-critical states of the search display (e.g., target presence/absence) and the repetition/switch of the target-defining dimension (color/ orientation) affect reaction time distributions. The results replicated the mean reaction time (RT) inter-trial and dimension repetition/switch effects that have been reported in previous studies. Going beyond this, to uncover the underlying mechanisms, we used the Drift-Diffusion Model (DDM) and the Linear Approach to Threshold with Ergodic Rate (LATER) model to explain the RT distributions in terms of decision bias (starting point) and information processing speed (evidence accumulation rate). We further investigated how these different aspects of the decision-making process are affected by different properties of stimulus history, giving rise to dissociable inter-trial effects. We approached this question by (i) combining each perceptual decision making model (DDM or LATER) with different updating models, each specifying a plausible rule for updating of either the starting point or the rate, based on stimulus history, and (ii) comparing every possible combination of trial-wise updating mechanism and perceptual decision model in a factorial model comparison. Consistently across experiments, we found that the (recent) history of the response-critical property influences the initial decision bias, while repetition/switch of the target-defining dimension affects the accumulation rate, likely reflecting an implicit 'top-down' modulation process. This provides strong evidence of a disassociation between response- and dimension-based inter-trial effects.
| S-EPMC6091979 | biostudies-literature