Dataset Information

Performance of Machine Learning Algorithms for Predicting Progression to Dementia in Memory Clinic Patients.

ABSTRACT:

Importance

Machine learning algorithms could be used as the basis for clinical decision-making aids to enhance clinical practice.

Objective

To assess the ability of machine learning algorithms to predict dementia incidence within 2 years compared with existing models and determine the optimal analytic approach and number of variables required.

Design, setting, and participants

This prognostic study used data from a prospective cohort of 15 307 participants without dementia at baseline to perform a secondary analysis of factors that could be used to predict dementia incidence. Participants attended National Alzheimer Coordinating Center memory clinics across the United States between 2005 and 2015. Analyses were conducted from March to May 2021.

Exposures

258 variables spanning domains of dementia-related clinical measures and risk factors.

Main outcomes and measures

The main outcome was incident all-cause dementia diagnosed within 2 years of baseline assessment.

Results

In a sample of 15 307 participants (mean [SD] age, 72.3 [9.8] years; 9129 [60%] women and 6178 [40%] men) without dementia at baseline, 1568 (10%) received a diagnosis of dementia within 2 years of their initial assessment. Compared with 2 existing models for dementia risk prediction (ie, Cardiovascular Risk Factors, Aging, and Incidence of Dementia Risk Score, and the Brief Dementia Screening Indicator), machine learning algorithms were superior in predicting incident all-cause dementia within 2 years. The gradient-boosted trees algorithm had a mean (SD) overall accuracy of 92% (1%), sensitivity of 0.45 (0.05), specificity of 0.97 (0.01), and area under the curve of 0.92 (0.01) using all 258 variables. Analysis of variable importance showed that only 6 variables were required for machine learning algorithms to achieve an accuracy of 91% and area under the curve of at least 0.89. Machine learning algorithms also identified up to 84% of participants who received an initial dementia diagnosis that was subsequently reversed to mild cognitive impairment or cognitively unimpaired, suggesting possible misdiagnosis.

Conclusions and relevance

These findings suggest that machine learning algorithms could accurately predict incident dementia within 2 years in patients receiving care at memory clinics using only 6 variables. These findings could be used to inform the development and validation of decision-making aids in memory clinics.

SUBMITTER: James C

PROVIDER: S-EPMC8678688 | biostudies-literature |

REPOSITORIES: biostudies-literature

ACCESS DATA

Similar Datasets

Project description:Background: The ability to assess adverse outcomes in patients with community-acquired pneumonia (CAP) could improve clinical decision-making to enhance clinical practice, but the studies remain insufficient, and similarly, few machine learning (ML) models have been developed. Objective: We aimed to explore the effectiveness of predicting adverse outcomes in CAP through ML models. Methods: A total of 2,302 adults with CAP who were prospectively recruited between January 2012 and March 2015 across three cities in South America were extracted from DryadData. After a 70:30 training set: test set split of the data, nine ML algorithms were executed and their diagnostic accuracy was measured mainly by the area under the curve (AUC). The nine ML algorithms included decision trees, random forests, extreme gradient boosting (XGBoost), support vector machines, Naïve Bayes, K-nearest neighbors, ridge regression, logistic regression without regularization, and neural networks. The adverse outcomes included hospital admission, mortality, ICU admission, and one-year post-enrollment status. Results: The XGBoost algorithm had the best performance in predicting hospital admission. Its AUC reached 0.921, and accuracy, precision, recall, and F1-score were better than those of other models. In the prediction of ICU admission, a model trained with the XGBoost algorithm showed the best performance with AUC 0.801. XGBoost algorithm also did a good job at predicting one-year post-enrollment status. The results of AUC, accuracy, precision, recall, and F1-score indicated the algorithm had high accuracy and precision. In addition, the best performance was seen by the neural network algorithm when predicting death (AUC 0.831). Conclusions: ML algorithms, particularly the XGBoost algorithm, were feasible and effective in predicting adverse outcomes of CAP patients. The ML models based on available common clinical features had great potential to guide individual treatment and subsequent clinical decisions.

Project description:BackgroundKidney disease progression rates vary among patients. Rapid and accurate prediction of kidney disease outcomes is crucial for disease management. In recent years, various prediction models using Machine Learning (ML) algorithms have been established in nephrology. However, their accuracy have been inconsistent. Therefore, we conducted a systematic review and meta-analysis to investigate the diagnostic accuracy of ML algorithms for kidney disease progression.MethodsWe searched PubMed, EMBASE, Cochrane Central Register of Controlled Trials, the Chinese Biomedicine Literature Database, Chinese National Knowledge Infrastructure, Wanfang Database, and the VIP Database for diagnostic studies on ML algorithms' accuracy in predicting kidney disease prognosis, from the establishment of these databases until October 2020. Two investigators independently evaluate study quality by QUADAS-2 tool and extracted data from single ML algorithm for data synthesis using the bivariate model and the hierarchical summary receiver operating characteristic (HSROC) model.ResultsFifteen studies were left after screening, only 6 studies were eligible for data synthesis. The sample size of these 6 studies was 12,534, and the kidney disease types could be divided into chronic kidney disease (CKD) and Immunoglobulin A Nephropathy, with 5 articles using end-stage renal diseases occurrence as the primary outcome. The main results indicated that the area under curve (AUC) of the HSROC was 0.87 (0.84-0.90) and ML algorithm exhibited a strong specificity, 95% confidence interval and heterogeneity (I2) of (0.87, 0.84-0.90, [I2 99.0%]) and a weak sensitivity of (0.68, 0.58-0.77, [I2 99.7%]) in predicting kidney disease deterioration. And the the results of subgroup analysis indicated that ML algorithm's AUC for predicting CKD prognosis was 0.82 (0.79-0.85), with the pool sensitivity of (0.64, 0.49-0.77, [I2 99.20%]) and pool specificity of (0.84, 0.74-0.91, [I2 99.84%]). The ML algorithm's AUC for predicting IgA nephropathy prognosis was 0.78 (0.74-0.81), with the pool sensitivity of (0.74, 0.71-0.77, [I2 7.10%]) and pool specificity of (0.93, 0.91-0.95, [I2 83.92%]).ConclusionTaking advantage of big data, ML algorithm-based prediction models have high accuracy in predicting kidney disease progression, we recommend ML algorithms as an auxiliary tool for clinicians to determine proper treatment and disease management strategies.

Project description:ObjectivesMachine learning algorithms are being increasingly used for predicting hospital readmissions. This meta-analysis evaluated the performance of logistic regression (LR) and machine learning (ML) models for the prediction of 30-day hospital readmission among patients in the US.MethodsElectronic databases (i.e., Medline, PubMed, and Embase) were searched from January 2015 to December 2019. Only studies in the English language were included. Two reviewers performed studies screening, quality appraisal, and data collection. The quality of the studies was assessed using the Quality in Prognosis Studies (QUIPS) tool. Model performance was evaluated using the Area Under the Curve (AUC). A random-effects meta-analysis was performed using STATA 16.ResultsNine studies were included based on the selection criteria. The most common ML techniques were tree-based methods such as boosting and random forest. Most of the studies had a low risk of bias (8/9). The AUC was greater with ML to predict 30-day all-cause hospital readmission compared with LR [Mean Difference (MD): 0.03; 95% Confidence Interval (CI) 0.01-0.05]. Subgroup analyses found that deep-learning methods had a better performance compared with LR (MD 0.06; 95% CI, 0.04-0.09), followed by neural networks (MD: 0.03; 95% CI, 0.03-0.03), while the AUCs of the tree-based (MD: 0.02; 95% CI -0.00-0.04) and kernel-based (MD: 0.02; 95% CI 0.02 (-0.13-0.16) methods were no different compared to LR. More than half of the studies evaluated heart failure-related rehospitalization (N = 5). For the readmission prediction among heart failure patients, ML performed better compared with LR, with a mean difference in AUC of 0.04 (95% CI, 0.01-0.07). The leave-one-out sensitivity analysis confirmed the robustness of the findings.ConclusionMultiple ML methods were used to predict 30-day all-cause hospital readmission. Performance varied across the ML methods, with deep-learning methods showing the best performance over the LR.

Project description:BACKGROUND:Determination of β-amyloid (Aβ) positivity and likelihood of underlying Alzheimer's disease (AD) relies on dichotomous biomarker cut-off values. Individuals with mild cognitive impairment (MCI) and Aβ within the normal range may still have a substantial risk of developing dementia, primarily of Alzheimer type. Their prognosis, as well as predictors of clinical progression, are not fully understood. The aim of this study was to explore the associations of cerebrospinal fluid (CSF) biomarkers (Aβ42, total tau, phosphorylated tau) and other characteristics, including modifiable vascular factors, with the risk of progression to dementia among patients with MCI and normal CSF Aβ42. METHODS:Three hundred eighteen memory clinic patients with CSF and clinical data, and at least 1-year follow-up, were included. Patients had normal CSF Aβ42 levels based on clinical cut-offs. Cox proportional hazard models with age as time scale and adjusted for sex, education, and cognition (Mini-Mental State Examination) were used to investigate predictors of progression to dementia and Alzheimer-type dementia. Potential predictors included CSF biomarkers, cognitive performance (verbal learning and memory), apolipoprotein E (APOE) ε4 genotype, medial temporal lobe atrophy, family history of dementia, depressive symptoms, and vascular factors, including the Cardiovascular Risk Factors, Aging and Dementia (CAIDE) risk score. Predictive performance of patient characteristics was further explored with Harrell C statistic. RESULTS:Lower normal Aβ42 and higher total tau and phosphorylated tau were associated with higher dementia risk, and the association was not driven by Aβ42 values close to cut-off. Additional predictors included poorer cognition, APOE ε4 genotype, higher systolic blood pressure, and lower body mass index, but not the CAIDE dementia risk score. Aβ42 individually and in combination with other CSF biomarkers improved the risk prediction compared to age and cognition alone. Medial temporal lobe atrophy or vascular factors did not increase the predictive performance. CONCLUSIONS:Possibility of underlying AD pathology and increased dementia risk should not be ruled out among MCI patients with CSF Aβ42 within the normal range. While cut-offs may be useful in clinical practice to identify high-risk individuals, personalized risk prediction tools incorporating continuous biomarkers may be preferable among individuals with intermediate risk. The role of modifiable vascular factors could be explored in this context.

Project description:Machine-learning models are susceptible to external influences which can result in performance deterioration. The aim of our study was to elucidate the impact of a sudden shift in covariates, like the one caused by the Covid-19 pandemic, on model performance. After ethical approval and registration in Clinical Trials (NCT04092933, initial release 17/09/2019), we developed different models for the prediction of perioperative mortality based on preoperative data: one for the pre-pandemic data period until March 2020, one including data before the pandemic and from the first wave until May 2020, and one that covers the complete period before and during the pandemic until October 2021. We applied XGBoost as well as a Deep Learning neural network (DL). Performance metrics of each model during the different pandemic phases were determined, and XGBoost models were analysed for changes in feature importance. XGBoost and DL provided similar performance on the pre-pandemic data with respect to area under receiver operating characteristic (AUROC, 0.951 vs. 0.942) and area under precision-recall curve (AUPR, 0.144 vs. 0.187). Validation in patient cohorts of the different pandemic waves showed high fluctuations in performance from both AUROC and AUPR for DL, whereas the XGBoost models seemed more stable. Change in variable frequencies with onset of the pandemic were visible in age, ASA score, and the higher proportion of emergency operations, among others. Age consistently showed the highest information gain. Models based on pre-pandemic data performed worse during the first pandemic wave (AUROC 0.914 for XGBoost and DL) whereas models augmented with data from the first wave lacked performance after the first wave (AUROC 0.907 for XGBoost and 0.747 for DL). The deterioration was also visible in AUPR, which worsened by over 50% in both XGBoost and DL in the first phase after re-training. A sudden shift in data impacts model performance. Re-training the model with updated data may cause degradation in predictive accuracy if the changes are only transient. Too early re-training should therefore be avoided, and close model surveillance is necessary.