No abstract

Systematic literature reviews provide rigorous assessments of clinical, cost-effectiveness, and humanistic data. Accordingly, there is a growing trend worldwide among healthcare agencies and decision-makers to require them in order to make informed decisions. Because these reviews are labor-intensive and time consuming, we applied advanced analytic methods (AAM) to determine if machine learning methods could classify abstracts as well as humans. Literature searches were run for metastatic non-small cell lung cancer treatments (mNSCLC) and metastatic castration-resistant prostate cancer (mCRPC) . Records were reviewed by humans and two AAMs. AAM-1 involved a pre-trained data-mining model specialized in biomedical literature, and AAM-2 was based on support vector machine algorithms. The AAMs assigned an accept/reject status, with reasons for exclusion. Automatic results were compared to those of humans. For mNSCLC, 5820 records were processed by humans and 440 (8%) records were accepted and the remaining items rejected. AAM-1 correctly accepted 6% of records and correctly excluded 79%. AAM-2 correctly accepted 6% of records and correctly excluded 82%. The review was completed by AAM-1 or AAM-2 in 52 hours, compared to 196 hours for humans. Work saved was estimated to be 76% and 79% by AAM-1 and AAM-2, respectively. For mCRPC, 2434 records were processed by humans and 26% of these were accepted and 74% rejected. AAM-1 correctly accepted 23% of records and rejected 62%. AAM-2 correctly accepted 20% of records and rejected 66%. The review was completed by AAM-1, AAM-2, and humans in 25, 25 and 85 hours, respectively. Work saved was estimated to be 61% and 68% by AAM-1 and AAM-2, respectively. AAMs can markedly reduce the time required for searching and triaging records during a systematic review. Methods similar to AAMs should be assessed in future research for how consistent their performances are in SLRs of economic, epidemiological and humanistic evidence.

BACKGROUND Hypertrophic cardiomyopathy (HCM) is a heart disease characterized by hypertrophy of the left ventricular myocardium. The disease is the most common cause of sudden cardiac death (SCD) in young people and competitive athletes due to fatal ventricular arrhythmias, but in most patients, however, HCM has a benign course. Therefore, it is of the utmost importance to properly evaluate patients and identify those who would benefit from a cardioverter-defibrillator (ICD) implantation. The HCM SCD-Risk Calculator is a useful tool for estimating the 5-year risk of SCD. Parameters included in the model at evaluation are: age, maximum left ventricular wall thickness, left atrial dimension, maximum gradient in left ventricular outflow tract, family history of SCD, non-sustained ventricular tachycardia and unexplained syncope. Patients’ risk of SCD is classified as low (<4%), intermediate (4-<6%) or high (≥6%). Those in the high-risk group should have an ICD implantation. It can also be considered in the intermediate-risk group. However, the calculator still needs improvement and machine learning (ML) has the potential to fulfill this task. ML algorithm creates a model for solving a specific problem without explicit programming - instead it relies only on available data - by discovering patterns and relations. METHODS 252 HCM patients (aged 20-88 years, 49,6% were men) treated in our Department from 2005 to 2018, have been enrolled. The follow-up lasted 0-13 years (average: 3.8 years). SCD was defined as sudden cardiac arrest (SCA) or an appropriate ICD intervention. All parameters from HCM SCD-Risk Calculator have been obtained and the risk of SCD has been calculated for all patients during the first echocardiographic evaluation. ML model with variables from HCM SCD-Risk Calculator has been created. Both methods have been compared. RESULTS 20 patients reached an SCD end-point. 1 patient died due to SCA and 19 had an appropriate ICD intervention. Among them, there were respectively 6, 7 and 7 patients in the low, intermediate and high-risk group of SCD. 1 patient, who died, had a low risk. The ML model correctly assessed the SCD event only in 1 patient. According to ML, the risk of SCD ≤2.07% was a negative predictor. CONCLUSIONS The study did not show an advantage of ML over HCM SCD-Risk Calculator. Because of the characteristic of the dataset (approximately the same number of features and observations), the selection of machine learning algorithms was limited. Best results (evaluated using LOOCV) were achieved with a decision tree. We expect that bigger dataset would allow improving model performance because of strong regularization need in the current setup.

No abstract