Abstracts / Annals of Epidemiology 24 (2014) 682e702 691

cross-classified mixed-effects model was fit, treating drugs as nested withindrug-classes, and patients and drugs as cross-classified.Results: Adherence varied within, but overlapped between drug-classes; forexample, average adherence to statins ranged from 64.8% (rosuvastatin) to75.1% (pravastatin), and beta-blockers from 63.8% (timolol) to 79.6% (nado-lol). Most variance in adherence was attributable to time-variant drug- andpatient-characteristics (65.9%), followed by time-invariant drug-specificattributes (29.2%); invariant patient characteristics and drug-class togetherexplained 4.9%. Substantial variance within drug-class remained afteradjusting for age, years living with diabetes, illness severity, number ofmedications, healthcare access, cost, socioeconomic status, sex, routes ofadministration, adverse events reported by drug, and combination-drugs.Conclusions: The majority of the variation in adherence is attributable tomodifiable factors, including the actual drug prescribed. Adjusted differencesin adherence by drug within classes may be used to select drugs with betteradherence profiles. Further research is needed to understand variation inadherence at the drug-level.

Infectious Disease

P37. Effects of Age, Period, and Cohort on Lung TuberculosisMortality Among Mexican Population from 1955 to 2012

Benjamin Acosta-Cazares MD, Nataly Tlecuitl-Mendoza MD. MexicanInstitute of the Social Security

Purpose: To describe temporal trends on lung tuberculosis mortality ratesand assess the effects of age, period and cohort on Mexican population.Methods: Cross sectional study. Age and sex-specific lung tuberculosismortality for the period of 1955-2012 were obtained from the World HealthOrganization. Deaths from lung tuberculosis were included according to theInternational Classification of Disease (ICD) 7th (A001 and B001), 8th (A006and B005), 9th (B020) and 10th (A15-A16). Adjusted Mortality rates per100,000 persons were calculated using 2010 Mexican population as thestandard. Amultivariate analysis to assess the effect of age, period and cohorton mortality was done by a Poisson regression assuming that the number ofdeaths follows a Poisson distribution and the death rates were a multi-plicative function of the included model parameters. For this analysis weused this model: log(dij/pij)¼Âm+?i+?j+?k. The effects of age, period andcohort were also evaluated by a graphical approach. Statistical analysis wascarried out using SAS version 9.3.Results: Lung tuberculosis mortality has decreased in the period 1955-2012inMexico; however it is higher in men. It is observed that mortality increaseswith age. Mortality in men has decreased from 34.23 cases per 100,000 in1955 to 2.32 in 2012. For women it was from 22.36 to 0.99. Our resultsshowed age and cohort effect in Poisson regression.Conclusions: Lung tuberculosis is still an important public health problem inMexico. Mortality has decreased in all age groups but there are age andcohort effects.

P38. Marginal Causal Effect of Year of Hospitalization on SepsisDiagnosis

S. Reza Jafarzadeh DVM, MPVM, PhD, Benjamin S. Thomas MD,Jonas Marschall MD, Victoria J. Fraser MD, Jeff Gill PhD, David K. WarrenMD, MPH. Washington University School of Medicine

Purpose: To quantify the coinciding improvement in the diagnosis anddocumentation of sepsis (hereafter ‘diagnosis’), as indicated by ICD-9-CMcoding in administrative data, by estimating changes in the probability ofdiagnosis in recent years.Methods: We conducted a retrospective cohort study of 98,267 hospital-izations in 66,208 unique patients who met systemic inflammatory responsesyndrome (SIRS) criteria at a tertiary-care center from 2008-2012. We esti-mated the marginal causal effect of year of hospitalization on diagnosis usinga g-computation approach, which uses a counterfactual framework relatingunobserved outcomes to hypothesized unobserved exposures, followingmixed-effects logistic regression modeling of sepsis diagnosis among SIRShospitalizations.Results: The observed frequency of sepsis diagnosis in SIRS hospitalizationswere 1,339 (6.8%), 1,401 (7.1%), 1,592 (8.0%), 1,789 (9.0%) and 1,994 (10.4%) inthe years from 2008 to 2012, respectively. In addition to year of

hospitalization, intensive care unit admission (odds ratio [OR] ¼ 5.5; 95%confidence interval [CI]: 4.8, 5.3) and number of blood cultures per hospi-talization (OR ¼ 1.3; 95% CI: 1.2, 1.3) were associated with diagnosis afteradjustment for demographics and Charlson comorbidity score. The causalrisk difference for diagnosis per 100 SIRS hospitalizations, had the hospi-talization occurred in 2012, was estimated to be 3.9% (95% CI: 3.8%, 4.0%),3.4% (95% CI: 3.3%, 3.5%), 2.2% (95% CI: 2.1%, 2.3%) and 0.9% (95% CI: 0.8%, 1.1%)from 2008-2011, respectively.Conclusions: In patients with identical risk factors and covariates, sepsisdiagnosis and its documentation improved significantly in recent years andcontributed to an apparent increase in sepsis incidence.

P39. Differences in Treatment of Chlamydia Trachomatis byAmbulatory Care Setting

William S. Pearson PhD, Thomas E. Gift PhD, Jami S. Leichliter PhD. Centersfor Disease Control and Prevention

Purpose: Chlamydia trachomatis (CT) is the most commonly reported sex-ually transmitted infection (STI) in the United States and timely, correcttreatment can reduce CT transmission and sequelae. Emergency depart-ments (ED) are an important location for diagnosing STIs. This study com-pared recommended treatment of CT in EDs to treatment in physician offices.Methods: Five years of data (2006-2010) were used from the NationalAmbulatory Care Survey (NAMCS), and the National Hospital AmbulatoryCare Surveys (NHAMCS), including the Outpatient survey (NHAMCS-OPD)and Emergency Department survey (NHAMCS-ED). All visits with a CTdiagnosis were selected for analysis. Among these visits, patient demo-graphics were reported, as well as whether a prescription for azithromycin ordoxycycline was written or given at the time of the visit. Differences inreceipt of recommended treatments were compared between visits tophysician offices and emergency departments using chi-square tests andlogistic regression models.Results: During the five year period, approximately 3.3 million ambulatorycare visits had diagnosed CT. A greater proportion of visits to EDs receivedthe recommended treatment compared to visits to physician offices (66.1% v.44.9%, p<.01). Controlling for patient’s age, sex and race/ethnicity, thosepresenting to the ED with CT were more likely to receive the recommendedantibiotic treatment as patients presenting to a physician’s office, (O.R. 2.18,95% C.I. 1.11 - 4.28).Conclusions: These analyses demonstrate both differences in the treatmentof CT by ambulatory care setting as well as opportunities for increasing use ofrecommended treatments for diagnosed cases of this important STI.

Injuries

P40. Risk Factors for Nonfatal Drowning in Children in RuralBangladesh: A Community-Based Case-Control Study

Syed Abul Hassan Md Abdullah MBBS, MPH(Epidemiology),Meerjady S. Flora MBBS, MPH, PhD. Bangladesh Army

Introduction: Most studies of drowning in Bangladeshi populations to datehave described mortality and trends. We sought to identify associationsbetween socioeconomic status and child-care practices and nonfataldrowning in rural Bangladeshi children.Methods: This community-based case-control study was conducted in ruralchildren aged 1-5 years in Raiganj subdistrict of Bangladesh. 122 cases and134 age matched controls were recruited and their mothers were inter-viewed by use of a structured questionnaire. Univariate analyses and logisticregression were done to analyse the data.Results: Child nonfatal drowning was significantly associated with mothers:With low educational status (P<0.001), younger age (P<0.005, of singlestatus (P<0.001)and with more than three children (P<0.001). Nonfataldrowning was 12 and five times more likely in children of illiterate mothersthan in children of mothers with academic knowledge equivalent to sixth totenth grade (OR [95% CI] 0.08[0.02-0.26]) and above tenth grade (OR[95% CI]0.21[0.04-0.95])(P<0.001), respectively. Low socioeconomic status, indicatedby lower family expenditure (P<0.001) and no house (P<0.05; OR[95% CI]0.58[0.17-0.99]), were found to be risk factors for childhood nonfataldrowning. Improved child care, as measured by a child care index, wasassociated with significantly lower nonfatal drowning (P<0.001). Child care