LARGE DATA BASE ANALYSIS AND HEALTH CARE UTILIZATION
* Finalists for the Lee B. Lusted Student Prize
The consistent empirical findings of a positive relationship between hospital surgical volume and outcome suggests a learning effect on the improvement of surgical practice – “practice makes perfect.”. The relationship provides a justification for concentrating surgical procedures in a limited number of high-volume hospitals (i.e., regionalization) as a way to improve surgical outcomes. In this study, we examine the degree of regionalization of 8 high-risk surgeries across states. Moreover, we optimize a statewide regionalization policy for those high-risk surgeries and assess its impact on in-hospital mortality and operating hospital capacity.
Our data are the 2003-2010 State Inpatient Database in 11 states. The 8 high-risk surgeries include repair of abdominal aortic aneurysm (AAA), aortic-valve replacement, coronary-artery bypass grafting (CABG), carotid endarterectomy, cystectomy, esophagectomy, lung resection, and pancreatectomy. For each surgery, we estimate a hospital-level learning curve to characterize the volume and in-hospital mortality relation. Next, using the curve, we solve best regionalization policy that minimizes in-hospital mortality. The gap between in-hospital mortality under existent patient referral pattern and that under best regionalization reflects regionalization degree of the surgery. We repeat the analysis across surgeries and states.
Regionalization of aortic-valve replacement varied the least across the 11 states (see Figure); its mortality gap between existent referral pattern and best regionalization ranged from 16.0% in New York state to 22.2% in California (6.2% difference). In contrast, regionalization of pancreatectomy varied the most across the 11 states (see Figure); its mortality gap between existent referral pattern and best regionalization ranged from 40.5% in Maryland to 66.5% in California (26.0% difference). Regionalization could avoid 16,719 deaths among 1,730,168 cases for the 8 surgeries in the 11 states during 2003-2010. Specifically, regionalization could avoid the most deaths (2,654 among 318,506 cases) for carotid endarterectomy and the least deaths (891 among 25,030 cases) for esophagectomy. Across the 8 surgeries, observed operating hospitals were 4.5 (for CABG) to 21.0 (for lung resection) times greater than optimal numbers under best regionalization.
Regionalization of high risk surgeries varies significantly in the US. Statewide regionalization has the potential to significantly reduce in-hospital mortality. From a mortality reduction perspective, there exists an overcapacity of caring hospitals for high risk surgeries.
Purpose: Previous research has shown that clinician response to FDA warnings may not be adequate, which could impact patient health negatively. Further, there may be variation in how clinicians respond to FDA warnings, which may lead to inconsistency in patient treatment and insufficient patient protection. While geographic variation in various aspects of healthcare in the United States has been documented widely, there is limited literature that documents geographic variation in response to FDA warnings. Understanding this variation is critical to policymakers and hospital administrators responsible for communicating FDA warnings and setting guidelines for its providers, particularly for large national healthcare systems, whose actions may be guided by the nature and extent of the observed variation.
Methods: We use a landmark FDA black box warning on a glucose lowering drug rosiglitazone to study geographic variation. Using a national cohort of 550,959 diabetes patients from the Department of Veterans Affairs (VA), we analyze variation in rosiglitazone use from 2003-2008, aggregated on a quarterly basis, for each of the 21 geographical regions in VA called the Veterans Integrated Service Networks (VISNs), a set of regional service networks that provide integrated care to veterans based on geographic location. We use multivariate logistic regression to estimate the effect of each VISN on the likelihood of a patient receiving a rosiglitazone prescription.
Results: At the aggregate VA level, rosiglitazone use decreased substantially after the FDA warnings. There was substantial geographical variation in rosiglitazone use before FDA warnings were issued, but this variation increased considerably afterwards. The variation factor rose from an average of 2.8 in the pre-warning periods to an average of 3.9 in the post-warning periods. The VISN in which the patient's primary care facility is located, significantly and differentially, affected the odds of a patient being prescribed rosiglitazone, after controlling for various patient and facility-level factors. Further, there are geographical differences in the timing of when rosiglitazone use reached its peak usage level in each VISN. Twelve out of 21 VISNs achieved their peak level before FDA warnings were issued.
Conclusion: Hospital administrators and policymakers need to eliminate barriers and develop effective mechanisms for disseminating information related to FDA warnings and setting up guidelines for its providers in order to reduce the variation.
Purpose: Multiple myeloma (MM) is the second most common hematologic malignancy in the United States and is preceded by monoclonal gammopathy of undetermined significance (MGUS). We aimed to investigate the association between obesity and the progression from MGUS to MM.
Methods: Patients with MGUS diagnosed between October 1, 1999 and December 31, 2009 were identified in the U.S. Veterans Health Administration (VHA) database based on the Ninth Revision of the International Classification of Diseases (ICD-9) code 273.1. Unique identifiers of patients were used to link data from the inpatient and outpatient data to the pharmacy data on MM treatment. MM incidence was determined by at least two occurrences of ICD-9 code 203.0 and treatment at any VHA facility within six months of diagnosis. Moreover, two investigators reviewed patient-level clinical data to verify actual diagnosis and date of diagnosis. Interval-censored survival analysis was used, because the time when MM occurs is not directly observed but is known to take place between MGUS and MM diagnoses. Nonparametric maximum likelihood estimator of the survival curves were generated using the expectation-maximization iterative convex minorant algorithm. Multivariate survival analysis, controlling for body mass index (BMI), gender, race, comorbidities, level of creatinine, marital status, and income level, was conducted by parametric accelerated failure time interval-censored analysis with Weibull-modeled survival time. BMI was categorized into normal-weight: 18.5≤BMI<25, overweight: 25≤BMI<30, and obese: BMI>30.
Results: Our sample comprised 9,430 MGUS patients. Among them, 509 patients (5.3%) progressed to MM. Survival curves show the patterns of transformation from MGUS to MM by BMI groups (Figure). In the multivariate analysis, overweight (HR: 1.48; 95% CI: 1.17-1.88) and obese (HR: 1.70; 95% CI: 1.32-2.19) patients were associated with an increased risk of transformation from MGUS to MM. Black race had a higher risk of progression (HR: 1.92; 95% CI: 1.58-2.35). Patients with higher creatinine level (HR: 0.74; 95% CI: 0.59-0.93) and higher Charlson comorbidity index (HR: 0.94; 95% CI: 0.92-0.98) were less likely to develop MM.
Conclusions: This study provides evidence suggesting that overweight or obesity was associated with an elevated risk of transformation from MGUS to MM. As obesity is the only modifiable risk factor for this transformation, we suggest that clinical practice recommend weight loss to people with higher risks of developing MGUS or MM.
Method: A random sample of 5000 ureterolithiasis patients was identified from administrative claims between 2008 through 2011. Diagnostic imaging and treatment interventions were extracted and analyzed in sequence analysis and then HCA. Obtained patterns were compared on service site, practitioner specialty, costs, imaging, and interventions.
Sequencing generated 167 patterns accounting for all patients and 25 sequences accounted for 90% of cases. The first three sequences (60%) were associated with CT scanning, pain medication in two, and medical expulsion therapy (MET) in one. Further analyses showed the first two sequences were associated with emergency department (ED) treatment (85%), while pattern 3 yielded no clear distinctions. No cost or other differences were observed.
Conversely, HCA retained 100% of patients in three interpretably meaningful practice patterns: watchful waiting (WW), complex-invasive treatment (CI), and urgent care (UC). The WW pattern was associated with outpatient treatment (45%) by primary care physicians (50%). The CI pattern was associated with both outpatient and inpatient service sites (46%) and Urologists (56%). The UC pattern was associated with ED (85%) and ER physicians (60%). WW showed the lowest frequency of CT imaging (65.6% versus CI (78.7%) and UC (99.8%)) (P < 0.001), the lowest use of MET (15.0% versus CI (23.1%) and UC (36.9%)) (P < 0.001). On average, CI cost $1,580 more per patient than WW and $1,359 more than UC (P < 0.05). UC cost $508 more per patient than WW (P < 0.05). CI was associated with higher use of ureteroscopy and shock wave lithotripsy treatments (P < 0.001).
Conclusion: HCA obtained three distinct practice patterns accounting for 100% of ureterolithiasis cases compared to sequencing, which required 167 distinct patterns to account for all patients. Furthermore, HCA provided interpretable and meaningful practice patterns; whereas, the patterns obtained from sequencing did not correspond with other external variables. Research aimed at describing practice patterns should consider HCA as an alternative to sequencing when appropriate.
Method: We obtained Medicare FFS data from a nationally representative 5% sample of Medicare beneficiaries and MA data from HealthCore’s Integrated Research Database (HIRD) including WellPoint affiliated BCBS plans. We compared imaging rates between Medicare FFS, where providers were subject to imaging payment reductions only, and MA, where providers were subject to both payment changes and radiology UM. The primary outcome was change in use of outpatient imaging in Ohio, Indiana, and Kentucky between the years 2008-2011. We also analyzed emergency room (ER) imaging use to examine possible substitution effect. We used propensity score weighting to reduce imbalances in observed covariates and applied regression analysis (including interaction of group and year effect) to compare changes in utilization between the cohorts.
Result: CT scans, MRIs and resting echocardiograms were the most common imaging tests in all years. After weighting, the rate of outpatient imaging declined from 700 per 1000 member-years in 2008 to 654 in 2011 in FFS, compared to a decline from 705 in 2008 to 603 in 2011 in MA, representing declines of 7% and 15%, respectively. Regression results showed that the UM program reduced utilization by 56 per 1000 enrollees (95% confidence interval, 38 to 74) in the MA group, relative to the FFS group. The ER imaging use increased for both cohorts; however the rate of change for each group was similar (p>0.05).
Conclusion: The presence of a UM program for imaging was associated with a significant reduction in outpatient imaging rates in MA enrollees as compared to FFS Medicare. This result was observed without increased imaging use in the ER setting relative to FFS. The program we evaluated is an example of how UM initiatives can complement national efforts such as Choosing Wisely and have greater impact as compared to efforts focused solely on payment reduction.
Method: This retrospective, observational study used administrative claims from the Truven Health MarketScan® Research databases. Patients met the following: administration of ADA, ETA, or UST between 2/8/2010-1/31/2011 (index date), ≥12 months continuous enrollment prior to and following the index date, 18+ years of age, and a diagnosis of PsO (ICD-9-CM diagnosis code 696.1x) prior to or on the index date. Patients with other immunologic disorders at baseline, including psoriatic arthritis, were excluded. Discontinuation was defined as a ≥90-day therapy gap following the end of days’ supply on an ADA or ETA claim or a ≥168-day gap following the end of days’ supply on a UST claim. Kaplan-Meier survival curves were created due to variable follow-up time. Patients were followed until discontinuation, disenrollment, or study end (1/31/2012), with censoring at disenrollment or 1/31/2012. Adjusted analyses using propensity score weighting on baseline variables were performed separately for UST vs ADA and UST vs ETA. The discontinuation hazard ratio (HR) was estimated using weighted Cox proportional hazard models. Restart, switch, time to restart and time to switch (all before discontinuation) were also examined.
Result: After weighting, a total of 2933 ADA, 4011 ETA, and 583 UST patients were included (mean age was 48.9-49.9 years and 55.1-58.2% male). UST patients had lower discontinuation rates compared to ADA or ETA patients (UST vs ADA: 39% vs 53%, p<0.001; UST vs ETA: 39% vs 56%, p<0.001) and similar mean time to discontinuation (UST vs ADA: 228 vs 231 days, p=0.815; UST vs ETA 228 vs 237 days, p=0.354). Overall, PsO patients were significantly less likely to discontinue from UST compared to ADA or ETA (UST vs ADA HR=0.626, p<0.0001; UST vs ETA HR=0.586, p<0.0001). Treatment re-starts were less frequent among UST patients (UST vs ADA: 9% vs 18%, p<0.001; UST vs ETA: 9% vs 23%, p<0.001). Fewer UST patients switched (UST vs ADA: 14% vs 21%, p<0.001; UST vs ETA: 15% vs 22%, p<0.001). Mean time to re-start ranged from 187-250 days. Mean time to switch ranged from 334-340 days.
Conclusion: Discontinuation, re-start, and switch rates were lower among PsO patient receiving UST compared to those receiving ADA or ETA.