Predicting Risk of Stroke From Lab Tests Using Machine Learning Algorithms: Development and Evaluation of Prediction Models

Background

Stroke, a cerebrovascular disease, is one of the major causes of death. It causes significant health and financial burdens for both patients and health care systems. One of the important risk factors for stroke is health-related behavior, which is becoming an increasingly important focus of prevention. Many machine learning models have been built to predict the risk of stroke or to automatically diagnose stroke, using predictors such as lifestyle factors or radiological imaging. However, there have been no models built using data from lab tests.

Objective

The aim of this study was to apply computational methods using machine learning techniques to predict stroke from lab test data.

Methods

We used the National Health and Nutrition Examination Survey data sets with three different data selection methods (ie, without data resampling, with data imputation, and with data resampling) to develop predictive models. We used four machine learning classifiers and six performance measures to evaluate the performance of the models.

Results

We found that accurate and sensitive machine learning models can be created to predict stroke from lab test data. Our results show that the data resampling approach performed the best compared to the other two data selection techniques. Prediction with the random forest algorithm, which was the best algorithm tested, achieved an accuracy, sensitivity, specificity, positive predictive value, negative predictive value, and area under the curve of 0.96, 0.97, 0.96, 0.75, 0.99, and 0.97, respectively, when all of the attributes were used.

Conclusions

The predictive model, built using data from lab tests, was easy to use and had high accuracy. In future studies, we aim to use data that reflect different types of stroke and to explore the data to build a prediction model for each type.

Prognostic Association between Common Laboratory Tests and Overall Survival in Elderly Men with De Novo Metastatic Castration Sensitive Prostate Cancer: A Population-Based Study in Canada

Simple Summary

Despite significant advancements in the treatment of metastatic prostate cancer, a validated prognostic tool for patients with de novo metastatic castration-sensitive prostate cancer (mCSPC) is still lacking. Using population-based data from Ontario, Canada, we examined the prognostic association between common laboratory tests and survival for patients with mCSPC. These low-cost commonly available laboratory tests, including neutrophil-to-lymphocyte ratio, platelet-to-lymphocyte ratio, albumin, hemoglobin, PSA decrease and PSA nadir, were significantly associated with OS. They can provide important prognostic information and should be utilized more frequently among patients with newly diagnosed mCSPC.

Abstract

De novo cases of metastatic prostate cancer (mCSPC) are associated with poorer prognosis. To assist in clinical decision-making, we aimed to determine the prognostic utility of commonly available laboratory-based markers with overall survival (OS). In a retrospective population-based study, a cohort of 3556 men aged ≥66 years diagnosed with de novo mCSPC between 2014 and 2019 was identified in Ontario (Canada) administrative database. OS was assessed by using the Kaplan–Meier method. Multivariate Cox regression analysis was performed to evaluate the association between laboratory markers and OS adjusting for patient and disease characteristics. Laboratory markers that were assessed include neutrophil-to-lymphocyte ratio (NLR), platelet-to-lymphocyte ratio (PLR), albumin, hemoglobin, serum testosterone and PSA kinetics. Among the 3556 older men with de novo mCSPC, their median age was 77 years (IQR: 71–83). The median survival was 18 months (IQR: 10–31). In multivariate analysis, a statistically significant association with OS was observed with all the markers (NLR, PLR, albumin, hemoglobin, PSA decrease, reaching PSA nadir and a 50% PSA decline), except for testosterone levels. Our findings support the use of markers of systemic inflammation (NLR, PLR and albumin), hemoglobin and PSA metrics as prognostic indicators for OS in de novo mCSPC.

Prognostic value of labs ordered on patients hospitalized with COVID-19

Background: Hospitalists at our institution have taken on most non-intensive care unit (ICU) coronavirus disease 2019 (COVID-19) care. Based on sparse research, our institution developed a protocol for ordering labs for this patient population, including routine admission labs in addition to eight COVID-19-specific daily labs. The study goal is to determine if COVID-19-specific admission labs have any prognostic value beyond that provided by routine admission labs and vitals, and costs of labs with no prognostic value.Methods: We retrospectively reviewed adult patients admitted with COVID-19 from 3/2020 to 7/2020. Outcomes were mortality, ICU stay, and length of hospitalization. Multivariable logistic and linear regression were used to determine if COVID-19-specific admission labs have any prognostic value beyond that provided by vitals and routine admission labs. COVID-19-specific labs were d-Dimer, fibrinogen, ferritin, LDH, CK, pro-BNP, troponin, and CRP. Multivariable models included all routine admission labs and vitals. COVID-19-specific admission labs were included in the multivariable models if the p-value was <0.05 in the univariable analysis.Results: 331 patients met study criteria, inpatient mortality was 13.0%, 52.4% of patients required ICU stays and the average length of hospitalization was 8.9 days. COVID-19-specific labs showed no additional prognostic value for mortality. CRP, LDH, and d-Dimer provided additional prognostic information for ICU stay. CRP≥100 mg/dL and LDH≥900 U/L were associated with increased length of hospitalization.Conclusion: Only 3 of 8 admission COVID-19-specific labs recommended by our institution's protocol had additional prognostic value beyond that provided by routine labs and vitals. The total cost of non-prognostic COVID-19-specific labs during the study period was $75,874.

Larvicidal Action of Cannabidiol Oil and Neem Oil against Three Stored Product Insect Pests: Effect on Survival Time and in Progeny

SStored product pests can be detrimental to agricultural produce. As much as chemical pesticides are effective control agents, they involve several environmental and health risks. Within the framework of studies on alternative pest management methods, interest has focused on a plethora of plants whose extracts have demonstrated promising action as insecticides. Azadirachta indica and the derived neem oil have been extensively tested against many harmful insect species. In contrast, Cannabis sativa L. and its main compound, CBD, a highly concentrated cannabinoid, have not been investigated much. The present study examined the potential insecticidal activity of CBD and neem oils against 4th instar larvae of Tribolium confusum, Oryzaephilus surinamensis and Plodia interpunctella on wheat, rice and corn seeds. Treatment efficacy was expressed in terms of larval mortality. Mortality was observed in relation to dosage, time exposure intervals and product types. The results showed clear pesticidal activity for both oils, which at high doses induced significant mortality. The treatments produced significantly fewer offspring in the insect species tested than the control. The efficacy of treatment in progeny suppression was, as expected, dose dependent.

Interventions to Educate Family Physicians to Change Test Ordering: Systematic Review of Randomized Controlled Trials

The purpose is to systematically review randomised controlled trials (RCTs) to change family physicians’ laboratory test-ordering. We searched 15 electronic databases (no language/date limitations). We identified 29 RCTs (4,111 physicians, 175,563 patients). Six studies specifically focused on reducing unnecessary tests, 23 on increasing screening tests. Using Cochrane methodology 48.5% of studies were low risk-of-bias for randomisation, 7% concealment of randomisation, 17% blinding of participants/personnel, 21% blinding outcome assessors, 27.5% attrition, 93% selective reporting. Only six studies were low risk for both randomisation and attrition. Twelve studies performed a power computation, three an intention-to-treat analysis and 13 statistically controlled clustering. Unweighted averages were computed to compare intervention/control groups for tests assessed by >5 studies. The results were that fourteen studies assessed lipids (average 10% more tests than control), 14 diabetes (average 8% > control), 5 cervical smears, 2 INR, one each thyroid, fecal occult-blood, cotinine, throat-swabs, testing after prescribing, and urine-cultures. Six studies aimed to decrease test groups (average decrease 18%), and two to increase test groups. Intervention strategies: one study used education (no change): two feedback (one 5% increase, one 27% desired decrease); eight education + feedback (average increase in desired direction >control 4.9%), ten system change (average increase 14.9%), one system change + feedback (increases 5-44%), three education + system change (average increase 6%), three education + system change + feedback (average 7.7% increase), one delayed testing. The conclusions are that only six RCTs were assessed at low risk of bias from both randomisation and attrition. Nevertheless, despite methodological shortcomings studies that found large changes (e.g. >20%) probably obtained real change.