Machine learning for identification of frailty in Canadian primary care practices

Abstract

IntroductionFrailty is a medical syndrome, commonly affecting people aged 65 years and over and is characterized by a greater risk of adverse outcomes following illness or injury. Electronic medical records contain a large amount of longitudinal data that can be used for primary care research. Machine learning can fully utilize this wide breadth of data for the detection of diseases and syndromes. The creation of a frailty case definition using machine learning may facilitate early intervention, inform advanced screening tests, and allow for surveillance. ObjectivesThe objective of this study was to develop a validated case definition of frailty for the primary care context, using machine learning. MethodsPhysicians participating in the Canadian Primary Care Sentinel Surveillance Network across Canada were asked to retrospectively identify the level of frailty present in a sample of their own patients (total n = 5,466), collected from 2015-2019. Frailty levels were dichotomized using a cut-off of 5. Extracted features included previously prescribed medications, billing codes, and other routinely collected primary care data. We used eight supervised machine learning algorithms, with performance assessed using a hold-out test set. A balanced training dataset was also created by oversampling. Sensitivity analyses considered two alternative dichotomization cut-offs. Model performance was evaluated using area under the receiver-operating characteristic curve, F1, accuracy, sensitivity, specificity, negative predictive value and positive predictive value. ResultsThe prevalence of frailty within our sample was 18.4%. Of the eight models developed to identify frail patients, an XGBoost model achieved the highest sensitivity (78.14%) and specificity (74.41%). The balanced training dataset did not improve classification performance. Sensitivity analyses did not show improved performance for cut-offs other than 5. ConclusionSupervised machine learning was able to create well performing classification models for frailty. Future research is needed to assess frailty inter-rater reliability, and link multiple data sources for frailty identification.

Machine Learning Approaches for the Frailty Screening: A Narrative Review

“…Table 1 provides an overview of the selected studies in the frailty screening for ML methods. In general, all studies classified frailty with only one tool, such as the Rockwood Clinical Frailty Scale (CFS) [ 36 ], electronic Frailty Index (eFI) [ 31 ]; frailty phenotype (FP) [ 37 ]; electronic Frailty Score (eFS) [ 5 ]; and an exception that utilized a combination of tools [ 21 ], which included FRS-26-ICD (frailty drawn from ICD-10 Clinical Modification), ECI (The Elixhauser Comorbidity Index (ECI), high-risk medications (10 risk classification, Beers Criteria, 2019), sociodemographic characteristics, healthcare, and insurance utilization. Another exception used a set of predictors variables, including clinical and socioeconomic aspects, and six target variables (mortality, disability, urgent hospitalization, fracture, preventable hospitalization, and accessing the emergency department with red code) [ 38 ].…”

“…After the removal of features with low variance or high correlation, they reduced the total number of features from 5466 to 75. They used DT, LR, SVM, NB, NN, KNN, RF, and XGBoost models, and the XGBoost was the model with the best results of the eight models which were developed; it achieved the highest sensitivity (78.14%) and specificity (74.41%), but the F1-score was not shown when they used the best threshold that was achieved by using the most optimal thresholds determined using ROC curves [ 36 ].…”

“…Regarding algorithms, SVM was the most often applied ML method for frailty screening [ 5 , 31 , 36 , 38 ]. SVMs are competent in finding the best possible separation of different categories by familiarizing the weights of polynomial functions.…”

“…These models use boosting, an ensemble method where each tree or model corrects errors made by earlier trees. XGBoost requires minimal feature engineering, allowing steps such as normalizations and scaling to be omitted, and outliers have little impact [ 21 , 35 , 36 ].…”

Identifying Frail Patients by Using Electronic Health Records in Primary Care: Current Status and Future Directions

“…However, in the primary care setting, Williamson et al used data from the Canadian Primary Care Sentinel Surveillance Network ( n = 875 adults aged over 65 years) to develop the EHR-based frailty definition by machine learning methods and showed that it had the poor predictive ability (sensitivity 28%, specificity 94%) compared with the CFS ( 80 ). However, another study based on the same database with a larger sample improved the predictive ability of this tool by using the XGBoost model (stands for eXtreme Gradient Boosting, is a gradient-boosted decision tree machine learning algorithm) and changing the decision threshold (sensitivity 78.14%, specificity 74.41%) ( 81 ). In addition, machine learning, such as the natural language processing algorithm, may help extract frailty variables from unstructured data in EHRs ( 82 ) and make full use of EHRs in primary care.…”