Avoiding Inferential Errors in Public Health Research: The Statistical Modelling of Physical Activity Behavior
Pages 384-391
Ann O. Amuta and Dudley Poston Jr
DOI: http://dx.doi.org/10.6000/1929-6029.2014.03.04.7
Published: 06 November 2014

Abstract:  Background: A review of the health behavior literature on the statistical modeling of days of physical activity (PA) indicates that in many instances linear regression models have been used. It is inappropriate statistically to model a count dependent variable such as days of physical activity with Ordinary Least Squares (OLS). Many count variables have skewed distributions, and, also, have a preponderance of zeroes. Count variables should not be treated as continuous and unbounded. If OLS is used, estimations of the regression will frequently turn out to be inefficient, inconsistent and biased, and such outcomes could well have incorrect impacts on health programs and policies.

Methods: We considered three statistical methods for modelling the distribution of days of PA data for respondents in the 2013 Health Information Trends Survey (HINTS). The three regression models analyzed were: Ordinary Least Squares (OLS), Negative Binomial (NBRM), and Zero-inflated Negative Binomial (ZINB). We used the exact same predictor variables in the three models. Our results illustrate the differences in the results.

Results: Our analyses of the PA data demonstrated that the ZINB model fits the observed PA data better than either the OLS or the NBRM models. The coefficients and standard errors differed in the zero-inflated count models from the other models. For instance, the ZINB coefficient for the association between income and PA behavior was not statistically significant (p>0.05), whereas in the NBRM and in the OLS models, it was statistically significant (p<0.05).

Conclusions: The inappropriate use of regression models could well lead to wrong statistical inferences. Our analyses of the number of days of moderate PA demonstrated that the ZINB count model fits the observed PA data much better than the OLS model and the NBRM.

Model Based Sparse Feature Extraction for Biomedical Signal Classification
Pages 34-48
Hilda Dhlakama and Siaka Lougue
DOI: https://doi.org/10.6000/1929-6029.2017.06.01.4

Published: 28 February 2017

Abstract: Background: Although the number of deaths has declined since 2007, Tuberculosis (TB) continues to be the number one cause of death in South Africa. To create a country free of TB, there is need for continued research to explore models that will provide the Department of Health with new interventions.

Aim: This study was aimed at identifying the risk factors of active self-reported TB prevalence for South Africa in 2014.

Methods: The Frequentist Logistic Regression (FLR) approach was applied on a sample of 19213 individuals taken from the National Income Dynamics Survey (NIDS) wave data. Bayesian analysis with non-informative priors were used to model Wave 1 to 3 data and elicitation of the obtained posterior density parameters by averaging done to obtain the informative priors used to model wave 4. The wave 4 results obtained under the two estimation approaches were compared as well as the results for non-informative and informative priors.

Results: The findings show that self-reported TB prevalence is higher than the reported 1%, Human Immuno Deficiency Virus (HIV) remains a major threat to TB and Eastern Cape is the province mostly affected by TB with Limpopo recording the least prevalence. Poor living conditions and lower socio-economic conditions continue to be drivers of TB whilst English illiteracy, lack of Secondary/Tertiary education, alcohol consumption, marital status, gender and age groups also influence TB progression to disease. FLR yielded similar results to Bayesian with non-informative priors whilst the results are more precise for informative priors.

Conclusion: This study identified individuals and communities at risk of developing active TB disease in South Africa.

Keywords: Tuberculosis, HIV, South Africa, National Income Dynamics Survey, Bayesian analysis, Frequentist Logistic Regression.

Download

Bayesian Inference Supports the Use of Bypass Surgery Over Percutaneous Coronary Intervention To Reduce Mortality in Diabetic Patients with Multivessel Coronary Disease
Pages 26-34
Christopher D. Lang, Yulei He and John A. Bittl
DOI: http://dx.doi.org/10.6000/1929-6029.2015.04.01.3
Published: 27 January 2015

Abstract: Background: Coronary artery bypass graft (CABG) surgery may confer a survival advantage over percutaneous coronary intervention (PCI) in diabetic patients with multivessel coronary artery disease (CAD), but results of individual studies have been mixed. The primary aim of the current study was to compare mortality rates in diabetic patients with multivessel CAD randomized to either or CABG or PCI at 5 years or longest follow-up.

Methods: Using a Bayesian approach, we updated a prior probability distribution elicited from 8 clinical trials (N=2024) with the likelihood obtained from the Future Revascularization Evaluation in Patients with Diabetes Mellitus: Optimal Management of Multivessel Disease (FREEDOM) (N=1460) to determine whether clinical trial evidence supports the underlying hypothesis that CABG is superior to PCI for diabetics with multivessel CAD.

Results:A conjugate normal model comparing mortality rates favored the use of CABG (posterior mean odds ratio [OR] = 0.58, 95% Bayesian credible interval [BCI] = 0.48–0.71). Models weighted by the use of drug-eluting stents also favored the use of CABG over PCI (OR = 0.61, 95% BCI 0.48–0.78), as did models weighted by study age (OR=0.64, 95% BCI 0.52–0.80) or use of arterial conduits (OR=0.64, 95% BCI 0.51–0.81). The results were supported by a Bayesian hierarchical meta-analysis using a non-informative prior distribution (OR=0.55, 95% BCI 0.37–0.76).

Conclusions: By integrating evidence from various studies, Bayesian methods directly support the underlying hypothesis that revascularization with CABG improves survival compared with PCI in diabetic patients with multivessel CAD.

Biologic Therapy for Psoriatic Arthritis or Moderate to Severe Plaque Psoriasis: Systematic Review with Pairwise and Network Meta-Analysis
Pages 74-87
Mariangela Peruzzi, Delia Colombo, Elena De Falco, Isotta Chimenti, Antonio Abbate, Giacomo Frati and Giuseppe Biondi-Zoccai
DOI: http://dx.doi.org/10.6000/1929-6029.2014.03.02.1
Published: 30 April 2014

Methods: Trials on biologic agents in psoriasis (including psoriatic arthritis) were sought in several databases. Endpoints were ≥75% Reduction in the Psoriasis Area and Severity Index (PASI75), ≥20% improvement in the American College of Rheumatology core set of outcomes (ACR20), serious adverse events (SAE), and adverse events (AE) at the longest available non-cross-over follow-up. Random-effect methods were used to obtain pairwise and network pooled estimates.

Results: A total of 52 trials with 17,617 patients and 9 different biologic agents included, with 52% affected by psoriatic arthritis. After an average follow-up of 18 weeks, treatment with placebo was associated with a 5.9% (5.2%-6.6%) rate of PASI75, 17.4% (15.1%-19.6%) of ACR20, 2.4% (1.9%-2.8%) of SAE, and 51.8% (50.2%-53.4%) of AE. Several biologic agents provided higher PASI75 rates than placebo, with golimumab yielding the most favorable results (relative risk [RR]=14.02 [6.85-17.11]). Accordingly, several agents provided higher ACR20 rates than placebo, with infliximab yielding the most favorable results (RR=3.02 [1.67-4.55]). Overall, rates of SAE and AE were higher for several but not all biologic agents versus placebo, with golimumab being associated with the most favorable results for SAE (RR=0.40 [0.11-1.41]), and abatacept for AE (RR=1.00 [0.79-1.22]).

Conclusions: Efficacy and safety of biologic agents for psoriasis differ, and clinicians should bear in mind these features to maximize safety and efficacy in the individual patient.

Keywords: Meta-analysis, Mixed treatment comparison, Network meta-analysis, Plaque psoriasis, Psoriasis, Psoriatic arthritis, Systematic review.