The adverse impact of ignoring multicollinearity on findings and data interpretation in regression analysis is very well documented in the statistical literature. The failure to identify and report multicollinearity could result in misleading interpretations of the results. A review of epidemiological literature in PubMed from January 2004 to December 2013, illustrated the need for a greater attention to identifying and minimizing the effect of multicollinearity in analysis of data from epidemiologic studies. We used simulated datasets and real life data from the Cameron County Hispanic Cohort to demonstrate the adverse effects of multicollinearity in the regression analysis and encourage researchers to consider the diagnostic for multicollinearity as one of the steps in regression analysis.

IntroductionDiabetes, hypertension, and hypercholesterolemia are common chronic diseases among Hispanics, a group projected to comprise 30% of the US population by 2050. Mexican Americans are the largest ethnically distinct subgroup among Hispanics. We assessed the prevalence of and risk factors for undiagnosed and untreated diabetes, hypertension, and hypercholesterolemia among Mexican Americans in Cameron County, Texas.MethodsWe analyzed cross-sectional baseline data collected from 2003 to 2008 in the Cameron County Hispanic Cohort, a randomly selected, community-recruited cohort of 2,000 Mexican American adults aged 18 or older, to assess prevalence of diabetes, hypertension, and hypercholesterolemia; to assess the extent to which these diseases had been previously diagnosed based on self-report; and to determine whether participants who self-reported having these diseases were receiving treatment. We also assessed social and economic factors associated with prevalence, diagnosis, and treatment.ResultsApproximately 70% of participants had 1 or more of the 3 chronic diseases studied. Of these, at least half had had 1 of these 3 diagnosed, and at least half of those who had had a disease diagnosed were not being treated. Having insurance coverage was positively associated with having the 3 diseases diagnosed and treated, as were higher income and education level.ConclusionsAlthough having insurance coverage is associated with receiving treatment, important social and cultural barriers remain. Failure to provide widespread preventive medicine at the primary care level will have costly consequences.

The application, timing, and duration of lockdown strategies during a pandemic remain poorly quantified with regards to expected public health outcomes. Previous projection models have reached conflicting conclusions about the effect of complete lockdowns on COVID-19 outcomes. We developed a stochastic continuous-time Markov chain (CTMC) model with eight states including the environment (SEAMHQRD-V), and derived a formula for the basic reproduction number, R0, for that model. Applying the $${R}_{0}$$ R 0 formula as a function in previously-published social contact matrices from 152 countries, we produced the distribution and four categories of possible $${R}_{0}$$ R 0 for the 152 countries and chose one country from each quarter as a representative for four social contact categories (Canada, China, Mexico, and Niger). The model was then used to predict the effects of lockdown timing in those four categories through the representative countries. The analysis for the effect of a lockdown was performed without the influence of the other control measures, like social distancing and mask wearing, to quantify its absolute effect. Hypothetical lockdown timing was shown to be the critical parameter in ameliorating pandemic peak incidence. More importantly, we found that well-timed lockdowns can split the peak of hospitalizations into two smaller distant peaks while extending the overall pandemic duration. The timing of lockdowns reveals that a “tunneling” effect on incidence can be achieved to bypass the peak and prevent pandemic caseloads from exceeding hospital capacity.

Globally half of all diabetes mellitus is undiagnosed. We sought to determine the extent and characteristics of undiagnosed type 2 diabetes mellitus and pre-diabetes in Mexican Americans residing in the United States. This disadvantaged population with 50% lifetime risk of diabetes is a microcosm of the current pandemic. We accessed baseline data between 2004 and 2014 from 2,838 adults recruited to our Cameron County Hispanic Cohort (CCHC); a two-stage randomly selected ‘Framingham-like’ cohort of Mexican Americans on the US Mexico border with severe health disparities. We examined prevalence, risk factors and metabolic health in diagnosed and undiagnosed diabetes and pre-diabetes. Two thirds of this Mexican American population has diabetes or pre-diabetes. Diabetes prevalence was 28.0%, nearly half undiagnosed, and pre-diabetes 31.6%. Mean BMI among those with diabetes was 33.5 kg/m2 compared with 29.0 kg/m2 for those without diabetes. Significant risk factors were low income and educational levels. Most with diabetes had increased waist/hip ratio. Lack of insurance and access to health services played a decisive role in failure to have diabetes diagnosed. Participants with undiagnosed diabetes and pre-diabetes had similar measures of poor metabolic health similar but generally not as severe as those with diagnosed diabetes. More than 50% of a minority Mexican American population in South Texas has diabetes or pre-diabetes and is metabolically unhealthy. Only a third of diabetes cases were diagnosed. Sustained efforts are imperative to identify, diagnose and treat individuals in underserved communities.

The application, timing, and duration of lockdown strategies during a pandemic remain poorly quantified with regards to expected public health outcomes. Previous projection models have reached conflicting conclusions about the effect of complete lockdowns on COVID-19 outcomes. We developed a stochastic continuous-time Markov chain (CTMC) model with eight states including the environment (SEAMHQRD-V), and derived a formula for the basic reproduction number, R0, for that model. Applying the R0 formula as a function in previously-published social contact matrices from 152 countries, we produced the distribution and four categories of possible R0 for the 152 countries and chose one country from each quarter as a representative for four social contact categories (Canada, China, Mexico, and Niger). The model was then used to predict the effects of lockdown timing in those four categories through the representative countries. The analysis for the effect of a lockdown was performed without the influence of the other control measures, like social distancing and mask wearing, to quantify its absolute effect. Hypothetical lockdown timing was shown to be the critical parameter in ameliorating pandemic peak incidence. More importantly, we found that well-timed lockdowns can split the peak of hospitalizations into two smaller distant peaks while extending the overall pandemic duration. The timing of lockdowns reveals that a “tunneling” effect on incidence can be achieved to bypass the peak and prevent pandemic caseloads from exceeding hospital capacity.