Characterizing the critical features when personalizing antihypertensive drugs using spectrum analysis and machine learning methods
Introduction
Hypertension is among the most common chronic diseases and is the most important risk factor in cardiovascular and cerebrovascular diseases. Major complications include stroke, myocardial infarction, heart failure, and chronic kidney disease, which lead to high mortality and a heavy burden on families and the society. The control of hypertension also has significance in reducing the incidence of hypertension-related cardiovascular disease hospitalizations and lowering the financial burden on the economy [1,2].
Medication is the most important treatment option for hypertension. The commonly used antihypertensive drugs belong to four categories: ACE inhibitors/ARBs, beta-blockers, calcium antagonists, and diuretics. A survey conducted by the International Collaborative Study of Cardiovascular Disease in ASIA (InterASIA) showed that only 8.1 % of hypertensive patients achieved blood pressure(BP) control (under 140/90 mmHg) [3]. Assessments of BP control around the world regularly show that 50 % of treated hypertensive patients reach their BP goals [4]. These surveys fully demonstrate that the control of hypertension is not satisfactory. Many aspects may contribute to the effect of antihypertensive drugs, such as individual differences in pathology and daily habits, psychological factors, and the intake of other medications. The final treatment effect, however, largely depends on the medication choice made by the doctor when prescribing the drugs. The need to individualize hypertension care has been documented in the Joint National Committee (JNC) Reports on the Detection, Evaluation, and Treatment of High Blood Pressure [5]. However, due to the complexity of influencing factors, it is very difficult for doctors to prescribe appropriate personalized prescriptions. The aim of this study was to characterize drug-related features from sizeable data to improve the personalization of antihypertensive agents.
The hospital information system (HIS) is a comprehensive, integrated information system used to record and manage all clinical information such as symptoms, laboratory examination results and clinical treatments for patients [6,7]. In recent years, due to the rapid development and widespread use of HIS, many clinical cases have accumulated, and the medical data mining technology has undergone considerable development in the healthcare domain to include diagnoses [8] and disease treatment [9]. For example, support vector machines (SVM) have contributed to predicting the difficulty of tracheal intubation in anesthesia [10]. The decision tree and k-nearest neighbor analysis are used to analyze the sensitive factors affecting the hospitalization of patients with pneumonia, which are highly supported by clinical medicine [11]. Machine learning-based methods are used in the classification of real-time tissue elastography for hepatic fibrosis in patients with chronic hepatitis B [12]. Artificial neural network is applied to predict the length of ICU stay after cardiac surgery [13]. However, due to the diversity of antihypertensive drugs and the individual differences of clinical samples, sizeable-data-based studies have seldom been conducted to determine the effective clinical features of these antihypertensive drugs, leading to a degree of uncertainty in the use of hypertension medications.
Taking advantage of the sizeable data, we aimed to apply data mining methods to reveal the clinical characteristics spectrum or important clinical attributes (sensitive factors) of five frequently used drugs (Irbesartan, Metoprolol, Felodipine, Amlodipine and Levamlodipine) for hypertension control. This was performed by comparing the successful and unsuccessful cases which provides a data-driven reference for the personalization of clinical antihypertensive drugs.
Section snippets
Data source
Information system of the West China Hospital of Sichuan University. The West China Hospital is a general hospital with more than 6000 beds. It serves both urban and rural citizens mainly in west China but not excluding other parts in China. The source includes all types of income groups.
Reference period: Jan2008 -Dec 2016.
Data population
Since there was no self-report system for hypertensive outpatients, we selected inpatients as our study object. All population whose diagnosis included the term
Average value
Average value is often a good indicator of group characteristics [24]. We grouped the patient cases according to the five most frequently used drugs and then calculated the average value for each feature before the initiation of treatment for each group (Table 3). From Table 3, the average values of the features show distinct differences for each drug.
Characteristic spectrum
Statistical analysis with visualization is the most intuitive way to express original data [25]. To achieve a better comparison of the overall
Discussion
Hypertension treatment remains perplexing because: (1) The evidence-based hypertension guideline management algorithm has established a goal for BP. The algorithm has also suggested categories for the available antihypertensive agents; however, a specific prescribed drug does not exist for patients with hypertension. In fact, doctors and patients only hope that the selected drug or therapeutic plan is efficacious. (2) For years, genomics experts have been striving to identify the most suitable
Conclusion
In this study, we analyzed 15 clinical features of five frequently used anti-hypertensive drugs by comparing successful and unsuccessful cases using statistical and machine learning-based methods. The differential characteristics spectrum for five drug groups were obtained and the weights of the features were calculated using the statistical and ensembled model. Our results showed that there is a difference between the calculated and decisive attributes that clinicians regularly use to diagnose
Sources of funding
This study has been financially supported by the Science and Technology Project of Sichuan Province (2015SZ0141) and the Natural Science Foundation of China (61773094). The funders had no role in the study design, in the collection, analysis and interpretation of data; in the writing of the manuscript; and in the decision to submit the manuscript for publication.
Acknowledgment
We would like to thank Editage (www.editage.com) for English language editing.
References (43)
Drug adherence in hypertension
Pharmacol Res
(2017)A review and comparison of classification algorithms for medical decision making
Health Policy (New York)
(2005)- et al.
SVM-based decision support system for clinic aided tracheal intubation predication with multiple features
Expert Syst Appl
(2009) - et al.
Machine-learning-based classification of real-time tissue elastography for hepatic fibrosis in patients with chronic hepatitis B
Comput Biol Med
(2017) - et al.
Use of a neural network as a predictive instrument for length of stay in the intensive care unit following cardiac surgery
Comput Biomed Res
(1993) - et al.
Performance analysis of data mining classification techniques to predict diabetes
Procedia Comput Sci
(2016) - et al.
A multilayer perceptron-based medical decision support system for heart disease diagnosis
Expert Syst Appl
(2006) - et al.
Serum cystatin C is superior to serum creatinine as a marker of kidney function: a meta-analysis
Am J Kidney Dis
(2002) - et al.
Amlodipine therapy corrects renal abnormalities encountered in the hypertensive state
Am J Kidney Dis
(1987) - et al.
Potential national and state medical care savings from primary disease prevention
Am J Public Health
(2011)
Antihypertensive medication adherence and subsequent healthcare utilization and costs
Am. J. Managed Care
Prevalence, awareness, treatment, and control ofHypertension in China
Hypertension
The national high blood pressure education program: 20 years ofAchievement
Description of a computerized adverse drug event monitor using a hospital information system
Hosp Pharm
Using a hospital information system to assess the effects of adverse drug events, Proceedings / the. Annual Symposium on Computer Application [sic] in Medical Care
Symp. Comput. Appl. Med. Care
Machine learning and data mining, methods and applications
Extract critical factors affecting the length of hospital of pneumonia patient by data mining (case study: an Iranian hospital)
Artif Intell Med
Evaluating the Framingham hypertension risk prediction model in young adults: the Coronary Artery Risk Development in Young Adults (CARDIA) study
Hypertension
Urinary potassium excretion and risk of developing hypertension: the prevention of renal and vascular end-stage disease study
Hypertension
Association of clinical and social factors with excess hypertension risk in black compared with White US adults
JAMA
Fatal and nonfatal outcomes, incidence of hypertension, and blood pressure changes in relation to urinary sodium excretion
JAMA
Cited by (0)
- 1
These authors contributed equally to this work.