Alternatives for logistic regression in cross-sectional studies: an empirical comparison of models that directly estimate the prevalence ratio

Barros, Aluísio J. D.; Hirakata, Vânia Naomi

doi:10.1186/1471-2288-3-21

Cited by 3,290 publications

(2,720 citation statements)

References 23 publications

Supporting

Mentioning

2,433

Contrasting

Unclassified

261

Order By: Relevance

“…We then compared the health status of PAH members to that of the general population sample by fitting robust Poisson regression models 31 to compute age-adjusted prevalence ratios (PR a ) of poor health status and their 95 % confidence intervals (95 % CI); we estimated PR a s overall and stratified by each explanatory variable. We obtained prevalence differences (PD) by subtracting the estimated age-standardized prevalence (Tables 2 and 3).…”

Section: Resultsmentioning

confidence: 99%

Foreclosure and Health in Southern Europe: Results from the Platform for People Affected by Mortgages

et al. 2016

View full text Add to dashboard Cite

Housing instability has been shown to be related to poorer health outcomes in various studies, mainly in the USA and UK. Affected individuals are more prone to psychiatric (e.g., major depression, anxiety) and physical disorders (e.g., hypertension). This situation has deteriorated with the onset of the economic crisis. One of the most affected countries is Spain, which has high rates of foreclosure and eviction that continue to rise. In response, a civil movement, The Platform for People Affected by Mortgages (PAH), works to provide solutions to its members affected by foreclosure and advocates for the right to decent housing. The aims of this study ware to describe and compare the health status of PAH members from Catalonia to a sample of the general population and to analyze the association between health status and mortgage status, foreclosure stage, and other socioeconomic variables, among members of the PAH. We conducted a cross-sectional study using a self-administered online questionnaire (2014) administered to 905 PAH members in Catalonia (918 years; 559 women and 346 men). Results were compared with health indicators from The Health Survey of Catalonia 2013 (n = 4830). The dependent variables were poor mental health (GHQ 12 ≥ 3), and poor self-reported health (fair or poor). All analyses were stratified by sex. We computed age-standardized prevalence and prevalence ratios of poor mental and self-reported health in both samples. We also analyzed health outcomes among PAH members according to mortgage status (mortgage holders or guarantors), stage of foreclosure, and other socioeconomic variables by computing prevalence ratios from robust Poisson regression models. The prevalence of poor mental health among PAH members was 90.6 % in women and 84.4 % in men, and 15.5 and 10.2 % in the general population, respectively. The prevalence of poor self-reported health was 55.6 % in women and 39.4 % in men from the PAH, and 19.2 and 16.1 % in the general population, respectively. These health inequalities were independent of socioeconomic status. The prevalence of poor mental health was higher among individuals in the non-payment stage of foreclosure than among those who were up to date with their payments (e.g., PR c = 1.16 [95 % CI 1.04-1.28]). In contrast, selfreported poor health was more prominent in later stages of foreclosure, such as in posteviction without dation in payment stage in men (PR c = 2.24 [95 %CI = 1.35-3.72]). We observed a considerably higher prevalence of poor mental and self-reported health among male and female PAH members than in the general population. Public policies that tackle housing instability and its consequences are urgently needed in Spain.

show abstract

Section: Resultsmentioning

confidence: 99%

Foreclosure and Health in Southern Europe: Results from the Platform for People Affected by Mortgages

et al. 2016

View full text Add to dashboard Cite

show abstract

“…Second, prevalence and 95% confidence intervals (95% CI) of the variables of interest were calculated. Given the cross‐sectional nature of the study, as well as potential high prevalence of the outcome of interest (Barros and Hirakata, 2003), the association between depression and glycemic control was evaluated using Poisson regression models with robust standard errors (Coutinho et al ., 2008), reporting prevalence ratios (PR) and 95% CI adjusted for potential confounders. Different models were fitted utilizing a hierarchical approach (Victora et al ., 1997), conducted to better understand the potential influence of covariates in the association of interest.…”

Section: Methodsmentioning

confidence: 99%

Association between depression and glycemic control among type 2 diabetes patients in Lima, Peru

Crispín‐Trebejo

Robles‐Cuadros

Bernabé‐Ortiz

2015

Asia-Pacific Psychiatry

View full text Add to dashboard Cite

IntroductionThere is limited and controversial information regarding the potential impact of depression on glycemic control. This study aims to evaluate the association between depression and poor glycemic control. In addition, the prevalence of depression and rates of poor glycemic control were determined.MethodsCross‐sectional study performed in the endocrinology unit of two hospitals of ESSALUD in Peru. The outcome of interest was poor glycemic control, evaluated by glycated hemoglobin (HbA1c: < 7% versus ≥ 7%), whereas the exposure of interest was depression defined as 15 or more points in the Patient Health Questionnaire‐9 tool.The association of interest was evaluated using Poisson regression models with robust standard errors reporting prevalence ratios (PR) and 95% confidence intervals (95% CI) adjusting for potential confounders.ResultsA total of 277 participants, 184 (66.4%) males, mean age 59.0 (SD: 4.8), and 7.1 (SD: 6.8) years of disease were analyzed. Only 31 participants (11.2%; 95% CI: 7.5%–14.9%) had moderately severe or severe depression, whereas 70 (25.3%; 95% CI 20.3%–30.8%) had good glycemic control. Depression increased the probability of having poor glycemic control (PR = 1.32; 95% CI 1.15–1.51) after adjusting for several potential confounders.ConclusionsThere is an association between depression and poor glycemic control among type 2 diabetes patients. Our results suggest that early detection of depression might be important to facilitate appropriate glycemic control and avoid further metabolic complications.

show abstract

“…When the estimates are on or near the boundaries of the valid parameter space, the estimation algorithm will not converge. The convergence problem is most likely to happen when the model contains a continuous or polychotomous covariate, or the response prevalence is high [13,14]. The simulation studies by Carter et al [11] show that the estimates have poor properties when the success probability approaches 1.…”

Section: The Estimation Methodsmentioning

confidence: 99%

“…As the actual response follows a binomial distribution, the variance of the coefficient β̂k tends to be overestimated [13]. Barros and Hirakata [13] and Zou [10] propose a robust variance estimator in the Poisson/Cox regression to adjust for over dispersion.…”

Section: The Estimation Methodsmentioning

confidence: 99%

See 1 more Smart Citation

Estimating relative risks for common outcome using PROC NLP

Yu¹,

Wang²

2008

Computer Methods and Programs in Biomedicine

View full text Add to dashboard Cite

In cross-sectional or cohort studies with binary outcomes, it is biologically interpretable and of interest to estimate the relative risk or prevalence ratio, especially when the response rates are not rare. Several methods have been used to estimate the relative risk, among which the log-binomial models yield the maximum likelihood estimate (MLE) of the parameters. Because of restrictions on the parameter space, the log-binomial models often run into convergence problems. Some remedies, e.g., the Poisson and Cox regressions, have been proposed. However, these methods may give outof-bound predicted response probabilities. In this paper, a new computation method using the SAS Nonlinear Programming (NLP) procedure is proposed to find the MLEs. The proposed NLP method was compared to the COPY method, a modified method to fit the log-binomial model. Issues in the implementation are discussed. For illustration, both methods were applied to data on the prevalence of microalbuminuria (micro-protein leakage into urine) for kidney disease patients from the Diabetes Control and Complications Trial. The sample SAS macro for calculating relative risk is provided in the appendix.

show abstract

Alternatives for logistic regression in cross-sectional studies: an empirical comparison of models that directly estimate the prevalence ratio

Cited by 3,290 publications

References 23 publications

Foreclosure and Health in Southern Europe: Results from the Platform for People Affected by Mortgages

Foreclosure and Health in Southern Europe: Results from the Platform for People Affected by Mortgages

Association between depression and glycemic control among type 2 diabetes patients in Lima, Peru

Estimating relative risks for common outcome using PROC NLP

Contact Info

Product

Resources

About