Application of Machine Learning to Identify Clustering of Cardiometabolic Risk Factors in U.S. Adults

Cardiometabolic and renal phenotypes and transitions in the United States population

Cardiovascular and renal conditions have both shared and distinct determinants. In this study, we applied unsupervised clustering to multiple rounds of the National Health and Nutrition Examination Survey from 1988 to 2018, and identified 10 cardiometabolic and renal phenotypes. These included a 'low risk' phenotype; two groups with average risk factor levels but different heights; one group with low body-mass index and high levels of high-density lipoprotein cholesterol; five phenotypes with high levels of one or two related risk factors ('high heart rate', 'high cholesterol', 'high blood pressure', 'severe obesity' and 'severe hyperglycemia'); and one phenotype with low diastolic blood pressure (DBP) and low estimated glomerular filtration rate (eGFR). Prevalence of the 'high blood pressure' and 'high cholesterol' phenotypes decreased over time, contrasted by a rise in the 'severe obesity' and 'low DBP, low eGFR' phenotypes. The cardiometabolic and renal traits of the US population have shifted from phenotypes with high blood pressure and cholesterol toward poor kidney function, hyperglycemia and severe obesity.

A Machine-Learning-Based Risk-Prediction Tool for HIV and Sexually Transmitted Infections Acquisition over the Next 12 Months

Background: More than one million people acquire sexually transmitted infections (STIs) every day globally. It is possible that predicting an individual’s future risk of HIV/STIs could contribute to behaviour change or improve testing. We developed a series of machine learning models and a subsequent risk-prediction tool for predicting the risk of HIV/STIs over the next 12 months. Methods: Our data included individuals who were re-tested at the clinic for HIV (65,043 consultations), syphilis (56,889 consultations), gonorrhoea (60,598 consultations), and chlamydia (63,529 consultations) after initial consultations at the largest public sexual health centre in Melbourne from 2 March 2015 to 31 December 2019. We used the receiver operating characteristic (AUC) curve to evaluate the model’s performance. The HIV/STI risk-prediction tool was delivered via a web application. Results: Our risk-prediction tool had an acceptable performance on the testing datasets for predicting HIV (AUC = 0.72), syphilis (AUC = 0.75), gonorrhoea (AUC = 0.73), and chlamydia (AUC = 0.67) acquisition. Conclusions: Using machine learning techniques, our risk-prediction tool has acceptable reliability in predicting HIV/STI acquisition over the next 12 months. This tool may be used on clinic websites or digital health platforms to form part of an intervention tool to increase testing or reduce future HIV/STI risk.

Predicting the diagnosis of HIV and sexually transmitted infections among men who have sex with men using machine learning approaches

OBJECTIVES

We aimed to develop machine learning models and evaluate their performance in predicting HIV and sexually transmitted infections (STIs) diagnosis based on a cohort of Australian men who have sex with men (MSM).

METHODS

We collected clinical records of 21,273 Australian MSM during 2011-2017. We compared accuracies for predicting HIV and STIs (syphilis, gonorrhoea, chlamydia) diagnosis using four machine learning approaches against a multivariable logistic regression (MLR) model.

RESULTS

Machine learning approaches consistently outperformed MLR. Gradient boosting machine (GBM) achieved the highest area under the receiver operator characteristic curve for HIV (76.3%) and STIs (syphilis, 85.8%; gonorrhoea, 75.5%; chlamydia, 68.0%), followed by extreme gradient boosting (71.1%, 82.2%, 70.3%, 66.4%), random forest (72.0%, 81.9%, 67.2%, 64.3%), deep learning (75.8%, 81.0%, 67.5%, 65.4%) and MLR (69.8%, 80.1%, 67.2%, 63.2%). GBM models demonstrated the ten greatest predictors collectively explained 62.7-73.6% of variations in predicting HIV/STIs. STIs symptoms, past syphilis infection, age, time living in Australia, frequency of condom use with casual male sexual partners during receptive anal sex and the number of casual male sexual partners in the past 12 months were most commonly identified predictors.

CONCLUSIONS

Machine learning approaches are advantageous over multivariable logistic regression models in predicting HIV/STIs diagnosis.

Cardiometabolic risk prediction algorithms for young people with psychosis: a systematic review and exploratory analysis

OBJECTIVE

Cardiometabolic risk prediction algorithms are common in clinical practice. Young people with psychosis are at high risk for developing cardiometabolic disorders. We aimed to examine whether existing cardiometabolic risk prediction algorithms are suitable for young people with psychosis.

METHODS

We conducted a systematic review and narrative synthesis of studies reporting the development and validation of cardiometabolic risk prediction algorithms for general or psychiatric populations. Furthermore, we used data from 505 participants with or at risk of psychosis at age 18 years in the ALSPAC birth cohort, to explore the performance of three algorithms (QDiabetes, QRISK3 and PRIMROSE) highlighted as potentially suitable. We repeated analyses after artificially increasing participant age to the mean age of the original algorithm studies to examine the impact of age on predictive performance.

RESULTS

We screened 7820 results, including 110 studies. All algorithms were developed in relatively older participants, and most were at high risk of bias. Three studies (QDiabetes, QRISK3 and PRIMROSE) featured psychiatric predictors. Age was more strongly weighted than other risk factors in each algorithm. In our exploratory analysis, calibration plots for all three algorithms implied a consistent systematic underprediction of cardiometabolic risk in the younger sample. After increasing participant age, calibration plots were markedly improved.

CONCLUSION

Existing cardiometabolic risk prediction algorithms cannot be recommended for young people with or at risk of psychosis. Existing algorithms may underpredict risk in young people, even in the face of other high-risk features. Recalibration of existing algorithms or a new tailored algorithm for the population is required.