Data source and sample

The Institutional Review Boards of the Central Texas Veterans Health Care System and of the VA Central Western Massachusetts Healthcare System approved the study using an expedited review and waiver of informed consent. The Scott & White Memorial Hospital Institutional Review Board exempted the study from oversight at Baylor Scott & White. Data were extracted from the STOPP data repository, a collection of administrative extracts from the VA’s all-electronic medical records system (EMR) on all patients treated during fiscal years 2006–2009 (FY2006-FY2009; Oct 2005-Sep 2009) [8] and from the VA’s Corporate Data Warehouse. Variables included dates and types of care, diagnoses, procedures, prescription medications, lab test results, outpatient care, and admission and inpatient stay details on 7 million veterans of US military service using the Veterans Health Administration for care. Administrative data are reported nightly from all facility EMR’s across the VA system and cumulated in the Corporate Data Warehouse and related files, accessible for research purposes via a secure VA intranet environment after IRB approval.

Inclusion criteria were age 18 years or older, veterans granted VA Priority 1–8 indicating eligibility for VA care based on military service or disability or poverty, valid triglycerides lab data, and admitted for AP during FY2006-FY2009. For patients with multiple qualifying admissions, their first during the observation period was used.

Measures

Variables assessed age, race, ethnicity, gender, marital status, and utilization (lab results; hospital admission; timing of care) as well as risk adjustment measures including mental and physical illnesses and the Charlson Comorbidity Index of conditions associated with post-discharge one-year mortality calculated without the two cancer measures (retained as separate indicators) [9,10].

The inclusion criterion for AP was defined by International Classification of Diseases 9th Edition (ICD9) diagnosis code 577.0 during an inpatient admission; chronic pancreatitis was defined by ICD9 code 577.1 (inpatient or outpatient diagnosis). The date of first AP admission during the observation period was the index date for each patient. Covariates included TG level, the maximum recorded lab result on the date closest to AP admission in the year prior to AP admission through the first week prior to admission. Lag between TG values and AP admission was determined in days, also categorized into quarters (9 to 12 months prior, 6 to 9 months prior, 3 to 6 months prior, and 0 to 3 months prior through 7 days pre-admission). TG values from -7 through +7 days were captured to examine hTG at the time of hospitalization for AP, in “month 0”. Based on American Heart Association guidelines published by Miller et al. 2011, TG values <150 mg/dL are categorized as normal, 150–199 mg/dL are borderline, 200–500 are very high [11]. However, the lack of evidence supporting the definition of 500+ as critically high led us to recategorize TG values for analysis as follows: <200 (normal/borderline), 200–499, 500–999, 1000–1999, 2000+; dichotomies of below and above 500, 1000, and 2000 were also created.

To ascertain and adjust for baseline comorbidity, the Charlson Comorbidity Index and other relevant disorders were assessed from International Classification of Diseases 9^th Edition (ICD-9) diagnoses recorded in the 365 days prior to AP admission. Conditions and the corresponding ICD-9 codes included the Charlson conditions of myocardial infarction (AMI; 410, 412), chronic heart failure (428), peripheral vascular disease, stroke (430–438), dementia (290), chronic obstructive pulmonary disease (490–496, 501–506.4), rheumatologic disease (710.0, 710.1, 710.4, 714.0, 714.1, 714.2, 714.81, 725), peptic ulcer (531–534), cirrhosis of the liver (571.2, 571.4, 571.5, 571.6), liver failure (572.2–572.8, 456.0, 456.1, 456.20, 456.21), diabetes (250), hemi/paraplegia (342, 344.1), chronic kidney disease (582–583, 585–586, 588), cancers (140–172, 174–195, 200–208), metastatic cancers (196, 197, 198, 199), AIDS (042–044); as well as CHD (414), gall bladder disease (574, 575, 576), osteoporosis (733), hypertension (401–405), hypothyroidism (244), and vitamin D deficiency (268). Additional conditions were assessed to understand variability in risk. Previous research has noted differences in post-hospitalization health outcomes for veterans with mental and behavioral disorders [8]. Alcohol disorders were indicated by ICD-9 diagnoses of 291, 303, or 305.0. Serious mental illnesses as defined by the VA [12] were major depressive disorder (MDD; 296.2, 296.3, 311), posttraumatic stress disorder (PTSD; 309.81), and psychotic disorders (bipolar disorder, 296.0–296.1, 296.4–296.8; and schizophrenia, 295). Prior-year AP admission during the 365 days before the index admission was identified from inpatient data. Prescription drug use was noted for statins, fibrates, and other lipid-lowering medications (LLRx) prescribed for at least 60 days in the 45–75 days prior to admission (see S1 Appendix).

Dependent variables

Adverse outcomes of acute pancreatitis include recurrent gall bladder disease and multiple organ failure and death; risks from uncontrolled lipids (primarily LDL-cholesterol) include coronary heart disease, stroke, and myocardial infarct. Outcomes in the 365 days after the index AP admission date therefore included gallbladder disease, cardiovascular disease (CVD) events and related procedures (CHD, stroke, MI, PCI, CABG), and death within 1 year post-AP (see S1 Appendix).

Analysis

To address the first aim, maximum TG level was graphed by month prior to AP admission to show temporal relationship of elevated TG with acute AP exacerbation. A Cox proportional hazards model estimated the association between TG level and time from TG assessment to AP admission adjusting for demographic and clinical covariates. To address the third aim regarding treatment with fibrates or statins, chi-square analysis assessed frequencies of and bivariate associations between receipt of fibrate, statins, or both and TG level. To address the second and fourth aims regarding adverse outcomes associated with LLRx patterns and outcomes following AP hospitalization, multivariable logistic regression modeled the association of LLRx treatment (fibrate-alone, statin+fibrate, statins-alone, neither fibrate nor statins and other LLRx (e.g., ezetimibe, niacin) and TG level with post-AP adverse events (gall bladder disorders, CHD, stroke, MI, PCI but not CABG as it was too rare to model), any of those clinical outcomes, one-year mortality, and the combined outcome (any adverse clinical outcome or death). The models were repeated on the subsample without alcohol or gall bladder disorders diagnosed pre- or post-AP.

Bivariate associations

Association between TG level and days until admission.

The number of days to AP admission from TG assessment varied significantly by TG level with shorter lag to admission among those with very high TG, ranging from 108.5 days for TG<200 to 90.9 days for TG 1000–1999 and 54.5 days for TG of 2000+ (F = 23.8; df = 4, 20603; p < .0001). Overall, TG level correlated modestly and negatively with days until admission for acute pancreatitis (r = -.07; p < .0001; see Fig 1), consistent with a temporal relationship between TG level and risk of AP assuming TG levels continued to increase between last lab test and AP admission. The adjusted Cox model estimated a small positive association between days from TG assay to AP admission and hTG (TG of 2000 mg/dL or more; HR = 1.38 95%CI 1.16–1.63) but not with lesser levels of elevated TG, in the range 200–1999 mg/dL.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Mean triglyceride level by months until admission for acute pancreatitis among 20,608 veterans (FY2006-FY2009).

https://doi.org/10.1371/journal.pone.0179998.g001

Multivariable models

Association of prescription pattern with adverse outcomes.

Controlling for demographics, prior-year acute pancreatitis, lag between TG assay and AP admission, Charlson comorbidity and cancers, alcohol and psychiatric disorders, hypertension, dyslipidemia, and prescribed LLRx, elevated triglyceride levels—as opposed to normal/borderline levels—were significantly inversely related to post-AP gall bladder disease, one-year mortality, CHD, and the combined clinical outcome as well as the combined clinical and mortality outcome; odds ratios ranged from 0.29 to 0.91 (see Table 2). Statins and fibrates were inversely associated with 1-year mortality, but were risk factors in the models of other clinical outcomes. Triglyceride level was not significantly related to the post-AP outcomes of stroke, MI, or PCI.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Factors associated with post-acute pancreatitis outcomes among 20,608 VA patients (FY2006-FY2009).

https://doi.org/10.1371/journal.pone.0179998.t002

Among those without alcohol or gall bladder disorders, risk of 1-year mortality was lower for those with elevated TG (results not shown). The models of MI, CHD, and stroke found no association with TG level. The combined cardiovascular outcome showed decreased risk associated with fibrates and increased risk associated with statins and other LLRx, at modest levels.