Using machine learning on cardiorespiratory fitness data for predicting hypertension: The Henry Ford ExercIse Testing (FIT) Project

This study evaluates and compares the performance of different machine learning techniques on predicting the individuals at risk of developing hypertension, and who are likely to benefit most from interventions, using the cardiorespiratory fitness data. The dataset of this study contains information of 23,095 patients who underwent clinician- referred exercise treadmill stress testing at Henry Ford Health Systems between 1991 and 2009 and had a complete 10-year follow-up. The variables of the dataset include information on vital signs, diagnosis and clinical laboratory measurements. Six machine learning techniques were investigated: LogitBoost (LB), Bayesian Network classifier (BN), Locally Weighted Naive Bayes (LWB), Artificial Neural Network (ANN), Support Vector Machine (SVM) and Random Tree Forest (RTF). Using different validation methods, the RTF model has shown the best performance (AUC = 0.93) and outperformed all other machine learning techniques examined in this study. The results have also shown that it is critical to carefully explore and evaluate the performance of the machine learning models using various model evaluation methods as the prediction accuracy can significantly differ.

The association of clinical indication for exercise stress testing with all-cause mortality: the FIT Project

INTRODUCTION

We hypothesized that the indication for stress testing provided by the referring physician would be an independent predictor of all-cause mortality.

MATERIAL AND METHODS

We studied 48,914 patients from The Henry Ford Exercise Testing Project (The FIT Project) without known congestive heart failure who were referred for a clinical treadmill stress test and followed for 11 ±4.7 years. The reason for stress test referral was abstracted from the clinical test order, and should be considered the primary concerning symptom or indication as stated by the ordering clinician. Hierarchical multivariable Cox proportional hazards regression was performed, after controlling for potential confounders including demographics, risk factors, and medication use as well as additional adjustment for exercise capacity in the final model.

RESULTS

A total of 67% of the patients were referred for chest pain, 12% for shortness of breath (SOB), 4% for palpitations, 3% for pre-operative evaluation, 6% for abnormal prior testing, and 7% for risk factors only. There were 6,211 total deaths during follow-up. Compared to chest pain, those referred for palpitations (HR = 0.72, 95% CI: 0.60-0.86) and risk factors only (HR = 0.72, 95% CI: 0.63-0.82) had a lower risk of all-cause mortality, whereas those referred for SOB (HR = 1.15, 95% CI: 1.07-1.23) and pre-operative evaluation (HR = 2.11, 95% CI: 1.94-2.30) had an increased risk. In subgroup analysis, referral for palpitations was protective only in those without coronary artery disease (CAD) (HR = 0.75, 95% CI: 0.62-0.90), while SOB increased mortality risk only in those with established CAD (HR = 1.25, 95% CI: 1.10-1.44).

CONCLUSIONS

The indication for stress testing is an independent predictor of mortality, showing an interaction with CAD status. Importantly, SOB may be associated with higher mortality risk than chest pain, particularly in patients with CAD.

Cardiorespiratory fitness and incident diabetes: the FIT (Henry Ford ExercIse Testing) project

OBJECTIVE

Prior evidence has linked higher cardiorespiratory fitness with a lower risk of diabetes in ambulatory populations. Using a demographically diverse study sample, we examined the association of fitness with incident diabetes in 46,979 patients from The Henry Ford ExercIse Testing (FIT) Project without diabetes at baseline.

RESEARCH DESIGN AND METHODS

Fitness was measured during a clinician-referred treadmill stress test performed between 1991 and 2009. Incident diabetes was defined as a new diagnosis of diabetes on three separate consecutive encounters derived from electronic medical records or administrative claims files. Analyses were performed with Cox proportional hazards models and were adjusted for diabetes risk factors.

RESULTS

The mean age was 53 years with 48% women and 27% black patients. Mean metabolic equivalents (METs) achieved was 9.5 (SD 3.0). During a median follow-up period of 5.2 years (interquartile range 2.6-8.3 years), there were 6,851 new diabetes cases (14.6%). After adjustment, patients achieving ≥12 METs had a 54% lower risk of incident diabetes compared with patients achieving <6 METs (hazard ratio 0.46 [95% CI 0.41, 0.51]; P-trend < 0.001). This relationship was preserved across strata of age, sex, race, obesity, hypertension, and hyperlipidemia.

CONCLUSIONS

These data demonstrate that higher fitness is associated with a lower risk of incident diabetes regardless of demographic characteristics and baseline risk factors. Future studies should examine the association between change in fitness over time and incident diabetes.

Prognostic value of exercise capacity in patients with coronary artery disease: the FIT (Henry Ford ExercIse Testing) project

OBJECTIVE

To examine the prognostic value of exercise capacity in patients with nonrevascularized and revascularized coronary artery disease (CAD) seen in routine clinical practice.

PATIENTS AND METHODS

We analyzed 9852 adults with known CAD (mean ± SD age, 61±12 years; 69% men [n=6836], 31% black race [n=3005]) from The Henry Ford ExercIse Testing (FIT) Project, a retrospective cohort study of patients who underwent physician-referred stress testing at a single health care system between January 1, 1991, and May 31, 2009. Patients were categorized by revascularization status (nonrevascularized, percutaneous coronary intervention [PCI], or coronary artery bypass graft [CABG] surgery) and by metabolic equivalents (METs) achieved on stress testing. Using Cox regression models, hazard ratios for mortality, myocardial infarction (MI), and downstream revascularizations were calculated after adjusting for potential confounders, including cardiac risk factors, pertinent medications, and stress testing indication.

RESULTS

There were 3824 all-cause deaths during median follow-up of 11.5 years. In addition, 1880 MIs, and 1930 revascularizations were ascertained. Each 1-MET increment in exercise capacity was associated with a hazard ratio (95% CI) of 0.87 (0.85-0.89), 0.87 (0.85-0.90), and 0.86 (0.84-0.89) for mortality; 0.98 (0.96-1.01), 0.88 (0.84-0.92), and 0.93 (0.90-0.97) for MI; and 0.94 (0.92-0.96), 0.91 (0.88-0.95), and 0.96 (0.92-0.99) for downstream revascularizations in the nonrevascularized, PCI, and CABG groups, respectively. In each MET category, the nonrevascularized group had similar mortality risk as and higher MI and downstream revascularization risk than the PCI and CABG surgery groups (P<.05).

CONCLUSION

Exercise capacity was a strong predictor of mortality, MI, and downstream revascularizations in this cohort. Furthermore, patients with similar exercise capacities had an equivalent mortality risk, irrespective of baseline revascularization status.

Leukocytosis in acute pulmonary embolism

PURPOSE

The purpose of this investigation is to assess the level of leukocytosis in acute pulmonary embolism (PE).

BACKGROUND

Limited data exist regarding leukocytosis in acute PE. One reason that the prevalence of leukocytosis in acute PE is unknown, despite an extensive number of investigations of PE, may relate to the fact that acute PE is usually associated with other conditions that themselves may cause leukocytosis.

METHODS

Hospital records of 386 patients with a diagnosis of acute PE were reviewed retrospectively. Patients with no other possible or definite cause of leukocytosis were analyzed separately. A diagnosis of PE was made by a high-probability interpretation of the ventilation/perfusion lung scan or pulmonary angiogram.

RESULTS

Among patients with PE in whom other possible or defined causes for leukocytosis were eliminated, 52 of 266 (20%) had a WBC count > 10,000/mm3. None had a WBC count that was > or = 20,000/mm3. Patients with the pulmonary hemorrhage/infarction syndrome had an increased WBC count in 32 of 183 (17%) vs 20 of 83 (24%) in patients who did not have pulmonary hemorrhage/infarction syndrome (not significant).

CONCLUSION

A modest leukocytosis may accompany (and possibly be caused by) PE. Its presence should not dissuade the clinician from objectively pursuing the diagnosis of PE.