Prevalence and major risk factors of non-communicable diseases: a machine learning based cross-sectional study

Abstract

The aim: The study aimed to determine the prevalence of several non-communicable diseases (NCD) and analyze risk factors among adult patients seeking nutritional guidance in Dhaka, Bangladesh. Participants: 146 hospitalized adults of both genders aged 18-93 participated in this cross-sectional research. Methods: We collected the demographic and vital information from 146 hospitalized patients in Dhaka, Bangladesh. We checked the physical and vital parameters, including blood sugar, serum creatinine, blood pressure, and the presence or absence of major non-communicable diseases. Then we used descriptive statistical approaches to explore the NCDs prevalence based on gender and age group. Afterwards, the relationship between different NCD pairs with their combined effects was analyzed using different hypothesis testing at a 95 % confidence level. Finally, the random forest and XGBoost machine learning algorithms are used to predict the comorbidity among the patients with the underlying responsible factors. Result: Our study observed the relationships between gender, age groups, obesity, and NCDs (DM, CKD, IBS, CVD, CRD, thyroid). The most frequently reported NCD was cardiovascular issues (CVD), which was present in 83.56 % of all participants. CVD was more common in male participants. Consequently, male participants had a higher blood pressure distribution than females. Diabetes mellitus (DM), on the other hand, did not have a gender-based inclination. Both CVD and DM had an age-based progression. Our study showed that chronic respiratory illness was more frequent in middle-aged participants than in younger or elderly individuals. Based on the data, every one in five hospitalized patients was obese. We analyzed the comorbidities and found that 31.5 % of the population has only one NCD, 30.1 % has two NCDs, and 38.3 % has more than two NCDs. Besides, 86.25 % of all diabetic patients had cardiovascular issues. All thyroid patients in our study had CVD. Using a t-test, we found a relationship between CKD and thyroid (p-value 0.061). Males under 35 years have a statistically significant relationship between thyroid and chronic respiratory diseases (p-value 0.018). We also found an association between DM and CKD among patients over 65 (p-value 0.038). Moreover, there has been a statistically significant relationship between CKD and Thyroid (P<0.05) for those below 35 and 35-65. We used a two-way ANOVA test to find the statistically significant interaction of heart issues and chronic respiratory illness in combination with diabetes. The combination of DM and RTI also affected CKD in male patients over 65 years old. Among machine learning algorithms, XGBoost produced the highest accuracy, 69.7 %, in comorbidity detection. Random forest feature importance detected age, weight and waist-hip ratio as the major risk factors behind the comorbidity. Conclusion: The prevalence study helps to identify the future risks and most vulnerable groups. By initiating and implementing control plans based on the prevalence study, it is possible to reduce the burden of NCDs on the elderly and middle-aged population of Bangladesh.