Michael L Pennell

The Ohio State University, Columbus, Ohio, United States

Are you Michael L Pennell?

Claim your profile

Publications (82)223.52 Total impact

  • [Show abstract] [Hide abstract]
    ABSTRACT: There is a lack of information about the correlates of completing all three cancer screening tests among women living in Appalachia. Cross-sectional telephone interviews were conducted (April-September 2013) among women (n=637) ages 51-75 from 12 Appalachia Ohio counties. Outcomes of within screening guidelines were verified by medical record. Multivariable logistic regression models identified correlates of being within guidelines for all three cancer screening tests. Screening rates were: mammography (32.1%), Pap test (36.1%), and a colorectal cancer test (30.1%). Only 8.6% of women were within guidelines for all tests. Having had a check-up in the past two years and having received a screening recommendation were significantly related to being within guidelines for all three tests (p<0.01). Participants with higher annual household incomes ($60,000+; OR=3.53, 95% CI: 1.49, 8.33) and conditions requiring regular medical visits (OR=3.16, 95% CI: 1.29, 7.74) were more likely to be within guidelines for all three screening tests. Less than 10% of women had completed screening within guidelines for all three screening tests. Regular contact with the healthcare system and higher incomes were significant predictors of being within guidelines. Within guidelines rates for the three recommended cancer screening tests is low among women in Appalachia Ohio. This finding illustrates the need for innovative interventions to improve rates of multiple cancer screening tests. Copyright © 2015, American Association for Cancer Research.
    Cancer Epidemiology Biomarkers & Prevention 08/2015; 24(10). DOI:10.1158/1055-9965.EPI-15-0369 · 4.13 Impact Factor
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: We hypothesized that the addition of toceranib to metronomic cyclophosphamide/piroxicam therapy would significantly improve disease-free interval (DFI) and overall survival (OS) in dogs with appendicular osteosarcoma (OSA) following amputation and carboplatin chemotherapy. This was a randomized, prospective clinical trial in which dogs with OSA free of gross metastatic disease (n = 126) received carboplatin chemotherapy (4 doses) following amputation. On study entry, dogs were randomized to receive piroxicam/cyclophosphamide with or without toceranib (n = 63 each) after completing chemotherapy. Patient demographics were not significantly different between both groups. During or immediately following carboplatin chemotherapy, 32 dogs (n = 13 toceranib; n = 19 control) developed metastatic disease, and 13 dogs left the study due to other medical conditions or owner preference. Following carboplatin chemotherapy, 81 dogs (n = 46 toceranib; n = 35 control) received the metronomic treatment; 35 dogs (n = 20 toceranib; n = 15 control) developed metastatic disease during the maintenance therapy, and 26 dogs left the study due to other medical conditions or owner preference. Nine toceranib-treated and 11 control dogs completed the study without evidence of metastatic disease 1-year following amputation. Toceranib-treated dogs experienced more episodes of diarrhea, neutropenia and weight loss than control dogs, although these toxicities were low-grade and typically resolved with supportive care. More toceranib-treated dogs (n = 8) were removed from the study for therapy-associated adverse events compared to control dogs (n = 1). The median DFI for control and toceranib treated dogs was 215 and 233 days, respectively (p = 0.274); the median OS for control and toceranib treated dogs was 242 and 318 days, respectively (p = 0.08). The one year survival rate for control dogs was 35% compared to 38% for dogs receiving toceranib. The addition of toceranib to metronomic piroxicam/cyclophosphamide therapy following amputation and carboplatin chemotherapy did not improve median DFI, OS or the 1-year survival rate in dogs with OSA.
    PLoS ONE 04/2015; 10(4):e0124889. DOI:10.1371/journal.pone.0124889 · 3.23 Impact Factor
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: To test the effectiveness of a colorectal cancer (CRC) screening intervention directed at three levels (clinic, provider, patient) in a primary care setting. We conducted a group randomized trial (Clinical Trials registration no. NCT01568151) among 10 primary care clinics in Columbus, Ohio that were randomized to a study condition (intervention or usual care). We determined the effect of a multi-level, stepped behavioral intervention on receipt of a CRC screening test among average-risk patients from these clinics over the study period. Patients (n=527) who were outside of CRC screening recommendations were recruited. Overall, 35.4% of participants in the intervention clinics had received CRC screening by the end of the study compared to 35.1% of participants who were in the usual care clinics. Time to CRC screening was also similar across arms (HR=0.97, 95% CI=0.65-1.45). The multi-level intervention was not effective in increasing CRC screening among participants who needed a test, perhaps due to low participation of patients in the stepped intervention. Future studies utilizing evidence-based strategies to encourage CRC screening are needed.
    04/2015; 2. DOI:10.1016/j.pmedr.2015.04.010
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: Objective. To test the effectiveness of a colorectal cancer (CRC) screening intervention among adults living in Ohio Appalachia. Methods. We conducted a group-randomized trial of a county-level intervention among adults living in 12 Ohio Appalachian counties who received a media campaign and clinic intervention focused on either CRC screening or fruits and vegetables. Participants' percentage within CRC screening guidelines was assessed with cross-sectional surveys conducted annually for four years, and validated with medical record review of screening. Results. On average, screening data were obtained on 564 intervention and 559 comparison participants per year. There was no difference in the Wave 4 CRC screening rates of intervention and comparison counties (35.2% versus 31.4%). Multivariate analyses found that high perceived risk of CRC, willingness to have a CRC test if recommended by a doctor, doctor recommendation of a CRC screening test, and patient-physician communication about changes in bowel habits, family history of CRC, and eating fruits and vegetables were significant (í µí± < 0.05) predictors of being within CRC screening guidelines. Conclusions. The intervention was not effective in increasing CRC rates among Ohio Appalachian adults. Future research should determine how media and clinic-based interventions can be modified to improve CRC screening rates among this underserved population.
    BioMed Research International 01/2015; 2015(4, article a18). DOI:10.1155/2015/943152 · 2.71 Impact Factor

  • Cancer Epidemiology Biomarkers & Prevention 11/2014; 23(11 Supplement):A42-A42. DOI:10.1158/1538-7755.DISP13-A42 · 4.13 Impact Factor

  • Cancer Research 10/2014; 74(19 Supplement):3809-3809. DOI:10.1158/1538-7445.AM2014-3809 · 9.33 Impact Factor
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: Background: The effect of neighborhood and healthcare access factors on cancer outcomes among patients enrolled in navigator programs is not clearly understood. This study assessed associations between: 1) neighborhood factors and diagnostic time to resolution (TTR) and, 2) geographic access and TTR following an abnormal breast or cervical cancer screening test among women participating in the Ohio Patient Navigator Research Program (OPNRP). Methods: Patient (demographic, socioeconomic status [SES], home-to-clinic distance) and neighborhood (deprivation, racial segregation) characteristics of 801 women living in one of 285 census tracts (CTs) in greater Columbus, Ohio were examined. Randomization to receive navigation occurred at the clinic level. Multilevel Cox regression and spatial analysis were used to estimate effects of various factors on TTR and assess model assumptions, respectively. Results: TTR increased as neighborhood deprivation increased. After adjustment for age, friend social support, education and healthcare status, the TTR among women living in a neighborhood with a moderate median household income (between $36,147 and $53,099) was shorter compared to women living in low median household income neighborhoods (< $36,147) (p < 0.05) . There is little evidence that unmeasured confounders are geographically patterned. Conclusions: Increased neighborhood socioeconomic deprivation was associated with longer TTR following an abnormal breast or cervical cancer screening test. Impact: These results highlight the need for addressing patient- and neighborhood-level factors to reduce cancer disparities among underserved populations.
    Cancer Epidemiology Biomarkers & Prevention 09/2014; 23(12). DOI:10.1158/1055-9965.EPI-14-0348 · 4.13 Impact Factor
  • [Show abstract] [Hide abstract]
    ABSTRACT: Self-efficacy (SE) has been found to be a robust predictor of success in achieving physical activity (PA) goals. While much of the current research has focused on SE as a trait, SE as a state has received less attention. Using day-to-day measurements obtained over 84 days, we examined the relationship between state SE and PA. Postmenopausal women (n = 71) participated in a 12-week PA intervention administered via cell phone and monitored their daily PA using a pedometer. At the end of each day, they reported their state SE and number of steps. Using a longitudinal model, state SE was found to be a robust predictor of PA even after accounting for trait SE and other covariates. The findings offer insights about the temporal relationship between SE and PA over the course of an intervention, which can be of interest to researchers and intervention designers.
    Health Education & Behavior 08/2014; 41(6). DOI:10.1177/1090198114543004 · 2.23 Impact Factor
  • [Show abstract] [Hide abstract]
    ABSTRACT: Abstract To date, no study has evaluated the short and long-term effects air pollution exposure on emphysematous subjects who have undergone lung volume reduction surgery (LVRS). Data from the NETT study (1998-2003) included 1218 subjects, aged 39-84. Daily values of ambient fine particulate matter (PM2.5) and ozone were obtained. Mixed effects models tested the association between short and long-term pollutant concentrations and changes in pulmonary function. Cumulative air pollution exposure was strongly associated with worsened respiratory function and symptoms. Mean PM2.5 was associated with poorer lung function. Lagged exposures were poorly associated with respiratory health outcomes. There were detrimental respiratory and pulmonary effects observed in response to even low levels of ambient air pollutants among study participants. These results are indicative that exposures even below those of air quality standards may still pose significant risks to severe COPD subjects.
    Archives of Environmental and Occupational Health 08/2014; 70(1). DOI:10.1080/19338244.2014.932753 · 0.93 Impact Factor
  • Source
    Roger Erich · Michael L Pennell ·
    [Show abstract] [Hide abstract]
    ABSTRACT: In this paper we propose a threshold regression (TR) model for time to event data related to subject health using a latent Ornstein-Uhlenbeck (OU) process that fails once it hits a boundary value for the first time. Baseline covariates are incorporated into the analysis using a log-link function for the initial state of the health process. The model provides clinically meaningful covariate effects and does not require the proportional hazards assumption of the commonly used Cox model. Unlike TR models based on the Wiener process, the OU model allows increments in the health process to depend on previous values and drifts toward a state of equilibrium or homeostasis, which are present in many biological applications. We also extend our model to incorporate a cure rate for applications with improper survival functions, such as time to tumor recurrence in a cancer clinical trial. Our models are applied to overall and relapse-free survival data of melanoma patients undergoing definitive surgery.
    Lifetime Data Analysis 08/2014; 21(1). DOI:10.1007/s10985-014-9306-8 · 0.65 Impact Factor
  • Jason P Richter · Ann Scheck McAlearney · Michael L Pennell ·
    [Show abstract] [Hide abstract]
    ABSTRACT: Although medical error reporting has been studied, underreporting remains pervasive. The study aims were to identify the organizational factors with the greatest perceived effect on error reporting and to determine whether associations differ for management and clinical staff. A total of 515 637 respondents from 1052 hospitals completed the Hospital Survey on Patient Safety Culture. Nine organizational factors were tested as predictors of error reporting using weighted least-squares multiple regression. Error feedback was perceived as the most significant predictor, while organizational learning was another significant factor. It also was found that although management support for patient safety was significantly related to error reporting among clinical staff, this association was not significant among management. This difference is relevant because managers may not be aware that their failure to demonstrate support for safety leads to underreporting by frontline clinical staff. Findings from this study can inform hospitals' efforts to increase error reporting.
    American Journal of Medical Quality 07/2014; 30(6). DOI:10.1177/1062860614544469 · 1.25 Impact Factor
  • [Show abstract] [Hide abstract]
    ABSTRACT: Background: Low-fat and low-carbohydrate weight-loss diets can have a beneficial effect on longitudinal measures of blood pressure and blood lipids. We aimed to assess longitudinal changes in blood pressure and blood lipids in a population of premenopausal women. We hypothesized that results may differ by level of adherence to the respective diet protocol and baseline presence of hypertension or hyperlipidemia. Methods: Overweight or obese premenopausal women were randomized to a low-fat (n=41) or low-carbohydrate (n=38) diet. As part of the 52-week Lifestyle Eating and Fitness (LEAF) intervention trial, we fit linear mixed models to determine whether a change in outcome differed by treatment arm. Results: Within-group trends in blood pressure and blood lipids did not differ (p>0.30). Across study arms, there was a significant decrease in systolic blood pressure (SBP, 3 mm Hg, p=0.01) over time, but diastolic blood pressure (DBP) did not change significantly over the course of the study. Blood lipids (total cholesterol [TC], low-density lipoproteins [LDL], and high-density lipoproteins [HDL]) all exhibited nonlinear trends over time (p<0.01); each decreased initially but returned to levels comparable to baseline by study conclusion (p>0.20). We observed a decline in SBP among women who were hypertensive at baseline (p<0.01), but hypercholesterolemia at baseline did not affect trends in blood lipids (p>0.40). Conclusions: Our results support that dietary interventions may be efficacious for lowering blood pressure and blood lipids among overweight or obese premenopausal women. However, a decrease in SBP was the only favorable change that was sustained in this study population. These changes can be maintained over the course of a 1-year intervention, yet changes in blood lipids may be less sustainable.
    Journal of Women's Health 07/2014; 23(8). DOI:10.1089/jwh.2013.4638 · 2.05 Impact Factor
  • Jason P Richter · Ann Scheck McAlearney · Michael L Pennell ·
    [Show abstract] [Hide abstract]
    ABSTRACT: Although patient handoffs have been extensively studied, they continue to be problematic. Studies have shown poor handoffs are associated with increased costs, morbidity, and mortality. No prior research compared perceptions of management and clinical staff regarding handoffs.
    Health care management review 07/2014; DOI:10.1097/HMR.0000000000000033 · 1.30 Impact Factor
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: Background Limited data are available on the association between colorectal cancer (CRC) worry and CRC screening uptake, particularly in rural and underserved populations where there is an excess burden of CRC. Methods Between September 2009 and March 2010, we conducted a cross-sectional study among a randomly selected sample of Appalachian Ohio residents aged 51-75 years (n= 1084). We also reviewed their medical records. Multivariable-adjusted models examined the association between CRC worry and screening by medical record review, assessed effect modification by CRC worry and determined the correlates of higher CRC worry. Results Approximately 50% of participants were adherent to CRC screening guidelines. There was no significant association between higher CRC worry and screening adherence [odds ratio (OR) = 1.32, 95% confidence interval (CI): 0.86-2.02]. CRC worry did not modify the association between any covariate and screening adherence. Participants who were unemployed/disabled (OR = 2.15, 95% CI: 1.34-3.45) and had higher CRC risk perception (OR = 3.49, 95% CI: 2.19-5.56) had higher odds of moderate-to-extreme worry. Conclusions These findings highlight the need for meaningful exploration of why higher CRC worry is not associated with adherence to CRC screening, particularly in rural, medically underserved populations. Development and implementation of interventions to increase CRC screening in such areas is a significant public health priority. © 2014 The Author 2014. Published by Oxford University Press on behalf of Faculty of Public Health. All rights reserved. For permissions, please e-mail: [email protected] /* */
    Journal of Public Health 05/2014; 37(2). DOI:10.1093/pubmed/fdu031 · 2.04 Impact Factor
  • [Show abstract] [Hide abstract]
    ABSTRACT: To compare exercise stress cardiac magnetic resonance (cardiac MR) to echocardiography in healthy volunteers with respect to adequacy of endocardial visualization and confidence of stress study interpretation. Twenty-eight healthy volunteers (age 28 ± 11 years, 15 males) underwent exercise stress echo and cardiac MR one week apart assigned randomly to one test first. Stress cardiac MR was performed using an MRI-compatible treadmill; stress echo was performed as per routine protocol. Cardiac MR and echo images were independently reviewed and scored for adequacy of endocardial visualization and confidence in interpretation of the stress study. Heart rate at the time of imaging was similar between the studies. Average time from cessation of exercise to start of imaging (21 vs. 31 s, P < 0.001) and time to acquire stress images (20 vs. 51 s, P < 0.001) was shorter for cardiac MR. The number of myocardial segments adequately visualized was significantly higher by cardiac MR at rest (99.8% vs. 96.4%, P = 0.002) and stress (99.8% vs. 94.1%, P = 0.001). The proportion of subjects in whom there was high confidence in the interpretation was higher for cardiac MR than echo (96% vs. 60%, P = 0.005). Exercise stress cardiac MR to assess peak exercise wall motion is feasible and can be performed at least as rapidly as stress echo. J. Magn. Reson. Imaging 2013. © 2013 Wiley Periodicals, Inc.
    Journal of Magnetic Resonance Imaging 05/2014; 39(5). DOI:10.1002/jmri.24263 · 3.21 Impact Factor
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: The most effective dietary pattern for breast cancer prevention has been greatly debated in recent years. Studies have examined hypocaloric diets, with particular emphasis on macronutrient composition, yielding inconclusive data. The objective of this study was to examine the effects of calorie-restricted low-fat and low-carbohydrate diets (LFD and LCD, respectively) on circulating adipokines among overweight and obese premenopausal women. Seventy-nine overweight and obese premenopausal women were randomized to either LFD or LCD, with increased physical activity, for 52 weeks. Serum adiponectin, leptin and the adiponectin-to-leptin ratio (A/L) were measured at baseline, and at weeks 34 and 52 to assess intervention effects. While there were no significant changes in serum adiponectin concentrations following the LCD and LFD interventions, leptin concentrations significantly decreased by week 34 of the intervention period (LCD: 35.3%, P = 0.004; LFD: 30.0%, P = 0.01), with no difference by intervention arm. At week 52, these reductions were statistically non-significant, indicating a return to baseline levels by the end of the intervention. While there were non-significant increases in the A/L ratio following the LCD and LFD intervention arms, the overall trend, across groups, was marginally significant (P = 0.05) with increases of 16.2% and 35.1% at weeks 34 and 52, respectively. These findings suggest that caloric-restricted LCD and LFD dietary patterns favorably modify leptin and possibly the A/L ratio, and lend support to the hypothesis that these interventions may be effective for obesity-related breast cancer prevention through their effects on biomarkers involved in metabolic pathways. Clinical Trial Registration Number: NCT01559194.
    SpringerPlus 04/2014; 3(1):175. DOI:10.1186/2193-1801-3-175
  • [Show abstract] [Hide abstract]
    ABSTRACT: Associations have been found between trusting patient-physician relationships and use of preventive care and a greater adherence to prescribed care. The objectives of this study were to assess the level of trust rural Medicaid smokers have in their physicians and whether trust was related to patient characteristics or physician behavior. This was a cross-sectional study of smokers who were enrolled in a tobacco-dependence treatment program. Participants were rural Medicaid-enrolled adults, age 18 and older, who were current smokers. Participants were enrolled from 8 primary care clinics as they came in for an appointment with their physician. The Trust in Physician Scale was completed at the baseline visit. One week later, an interview was conducted with the smoker to determine whether the physician provided tobacco-dependence treatment counseling at the visit. Mixed models were used to model the relationship between trust and participant characteristics and physician behaviors. Medicaid smokers in this study exhibited a high level of trust in their health care provider, as levels were similar to those reported in the general population of patients. Trust was significantly higher among individuals with better self-reported health. Rural Medicaid smokers appeared to have similar levels of trust in their physician as other patients. Future research should explore the role trust plays in shaping interactions between underserved populations and physicians within the context of smoking cessation counseling.
    The Journal of Rural Health 04/2014; 30(2):214-20. DOI:10.1111/jrh.12046 · 1.45 Impact Factor
  • Beom Seuk Hwang · Michael L Pennell ·
    [Show abstract] [Hide abstract]
    ABSTRACT: Many dose-response studies collect data on correlated outcomes. For example, in developmental toxicity studies, uterine weight and presence of malformed pups are measured on the same dam. Joint modeling can result in more efficient inferences than independent models for each outcome. Most methods for joint modeling assume standard parametric response distributions. However, in toxicity studies, it is possible that response distributions vary in location and shape with dose, which may not be easily captured by standard models. To address this issue, we propose a semiparametric Bayesian joint model for a binary and continuous response. In our model, a kernel stick-breaking process prior is assigned to the distribution of a random effect shared across outcomes, which allows flexible changes in distribution shape with dose shared across outcomes. The model also includes outcome-specific fixed effects to allow different location effects. In simulation studies, we found that the proposed model provides accurate estimates of toxicological risk when the data do not satisfy assumptions of standard parametric models. We apply our method to data from a developmental toxicity study of ethylene glycol diethyl ether. Copyright © 2013 John Wiley & Sons, Ltd.
    Statistics in Medicine 03/2014; 33(7). DOI:10.1002/sim.6007 · 1.83 Impact Factor
  • [Show abstract] [Hide abstract]
    ABSTRACT: Iron has been implicated in atherogenesis and plaque destabilization, whereas less is known about iron-related proteins in this disease. We compared ex vivo quantities with in vivo vessel wall T2*, which is a noncontrast magnetic resonance relaxation time that quantitatively shortens with increased tissue iron content. We also tested the hypothesis that patients with carotid atherosclerosis have abnormal T2* times vs controls that would help support a role for iron in human atherosclerosis. Forty-six patients undergoing carotid endarterectomy and 14 subjects without carotid disease were prospectively enrolled to undergo carotid magnetic resonance imaging. Ex vivo measurements were performed on explanted plaque and 17 mammary artery samples. Plaques vs normal arteries had higher levels of ferritin (median, 7.3 [interquartile range (IQR), 4-13.8] vs 1.0 [IQR, 0.6-1.3] ng/mg; P < .001) and oxidized low-density lipoprotein (median, 0.17 [IQR, 0.12-0.30] vs 0.01 [IQR, 0.003-0.03] ng/mg; P < .001) as well as hepcidin (median, 8.7 [IQR, 4.6-12.4] vs 2.6 [IQR, 1.3-7.0] ng/mL; P = .03); serum hepcidin levels did not distinguish atherosclerosis patients from controls (median, 40.6 [IQR, 18.8-88.6] vs 33.9 [IQR, 17.6-55.2]; P = .42). Shorter in vivo T2* paralleled larger plaque volume (ρ = -.44; P = .01), and diseased arteries had shorter T2* values compared with controls (median, 17.7 ± 4.3 vs 23.0 ± 2.4 ms; P < .001). Diseased arteries have greater levels of iron-related proteins ex vivo and shorter T2* times in vivo. Further studies should help define the role of T2* as a biomarker of iron and atherosclerosis.
    Journal of vascular surgery: official publication, the Society for Vascular Surgery [and] International Society for Cardiovascular Surgery, North American Chapter 03/2014; 61(6). DOI:10.1016/j.jvs.2014.02.006 · 3.02 Impact Factor
  • J Plascak · Aa Llanos · M Pennell · Rc Weier · Ed Paskett ·
    [Show abstract] [Hide abstract]
    ABSTRACT: This study's aim was to assess the associations between neighborhood and geographic factors and diagnostic resolution within 12 months following an abnormal screening test for breast or cervical cancer among participants of the Ohio Patient Navigator Research Program. Methods: Patient (demographic, psychosocial, and socioeconomic status [SES]) and neighborhood (SES, racial segregation, healthcare access) factors of 776 women attending one of 16 clinics and living in one of 285 census tracts in greater Columbus, Ohio were examined. Women were randomized at the clinic level to determine receipt of the navigation intervention. Multilevel logistic regression was used to estimate associations between these factors and diagnostic resolution while controlling for clinic and neighborhood clustering. Spatial analysis assessed the presence of geographic disparities (i.e., regions of significantly high odds of not resolving). Results: After adjustment for individual-level SES and the receipt of the clinic-level navigation intervention, there was a significant inverse association between neighborhood-level percent of residents without a bachelor's degree and odds of diagnostic resolution (odds ratio = 0.79, 95% confidence interval = 0.63-0.98). Adjustment for neighborhood percent of residents without a bachelor's degree accounted for geographic disparities in diagnostic resolution. Racial segregation, home-to-clinic distance and PN were not associated with odds of resolution. Conclusions: Women were more likely to have diagnostically resolved 12 months following an abnormal breast or cervical cancer screening test if they lived in neighborhoods of higher SES. Geographic disparities in diagnostic resolution could be reduced if there was no variation in neighborhood SES.
    Cancer Epidemiology Biomarkers & Prevention 03/2014; 23(3):569. DOI:10.1158/1055-9965.EPI-14-0080 · 4.13 Impact Factor

Publication Stats

807 Citations
223.52 Total Impact Points


  • 2008-2015
    • The Ohio State University
      • • Division of Biostatistics
      • • Center for Biostatistics
      Columbus, Ohio, United States
  • 2005-2007
    • University of North Carolina at Chapel Hill
      • Department of Biostatistics
      Chapel Hill, NC, United States
  • 2006
    • National Institutes of Health
      • Branch of Epidemiology (EPI)
      Bethesda, MD, United States