Logistic regression vs. linear regression in exploring the importance of health literacy for the quality of life of patients with heart failure

Abstract Methods A cross sectional study was conducted among 200 heart failure patients selected from two hospitals in Alexandria governate. The HLS-EU-Q16 survey to assess HL, SF-12 health survey questionnaire to assess QoL, record review, and interviewing the patients were used to collect the required data. Linear and logistic regression models were used to explore the relation between health literacy and the QoL then comparing between the two models. Results Forty-six and half percent of HF patients had sufficient HL while 53.5% had insufficient HL. Patients showed low mean score of 34.16 ± 10.43 for the physical dimension of the SF-12 questionnaire (PCS-12), while the mental dimension (MCS-12) displayed good mean score of 41.87 ± 14.34. For the linear regression regarding PCS-12, HL was statistically significant and an independent predictor of PCS-12 in patients with HF, both before and after adjustment for the personal and clinical characteristics. Concerning MCS-12, we found that HL was statistically significant and an independent predictor of MCS-12 only before adjustment. While in logistic regression, considering PCS-12, analysis revealed that HL was not statistically significant predictor of the median-split dichotomized PCS-12 in both before and after adjustments for covariates. Regarding MCS-12, it was found that HL was statistically significant and an independent predictor of the median-split dichotomized MCS-12 only before adjustment for personal and clinical characteristics. The correct prediction percent in the logistic model of PCS-12 (70%) was higher than the linear model (53.5%). Similarly, the correct prediction percent in the logistic model of MCS-12 (68.5%) was higher than the linear model (57.5%). Conclusion It has been explained that HL is a statistically significant independent predictor of only the PCS-12 of SF-12 in the linear regression model but not for MCS-12, while it was not statistically significant predictor in the logistic models for the two dimensions of SF-12. However, each of the two models, linear regression model and logistic regression model, has its advantages and disadvantages.