Data resource profile: State Inpatient Databases

Evaluating Algorithmic Bias in 30-Day Hospital Readmission Models: Retrospective Analysis

BACKGROUND

The adoption of predictive algorithms in health care comes with the potential for algorithmic bias, which could exacerbate existing disparities. Fairness metrics have been proposed to measure algorithmic bias, but their application to real-world tasks is limited.

OBJECTIVE

This study aims to evaluate the algorithmic bias associated with the application of common 30-day hospital readmission models and assess the usefulness and interpretability of selected fairness metrics.

METHODS

We used 10.6 million adult inpatient discharges from Maryland and Florida from 2016 to 2019 in this retrospective study. Models predicting 30-day hospital readmissions were evaluated: LACE Index, modified HOSPITAL score, and modified Centers for Medicare & Medicaid Services (CMS) readmission measure, which were applied as-is (using existing coefficients) and retrained (recalibrated with 50% of the data). Predictive performances and bias measures were evaluated for all, between Black and White populations, and between low- and other-income groups. Bias measures included the parity of false negative rate (FNR), false positive rate (FPR), 0-1 loss, and generalized entropy index. Racial bias represented by FNR and FPR differences was stratified to explore shifts in algorithmic bias in different populations.

RESULTS

The retrained CMS model demonstrated the best predictive performance (area under the curve: 0.74 in Maryland and 0.68-0.70 in Florida), and the modified HOSPITAL score demonstrated the best calibration (Brier score: 0.16-0.19 in Maryland and 0.19-0.21 in Florida). Calibration was better in White (compared to Black) populations and other-income (compared to low-income) groups, and the area under the curve was higher or similar in the Black (compared to White) populations. The retrained CMS and modified HOSPITAL score had the lowest racial and income bias in Maryland. In Florida, both of these models overall had the lowest income bias and the modified HOSPITAL score showed the lowest racial bias. In both states, the White and higher-income populations showed a higher FNR, while the Black and low-income populations resulted in a higher FPR and a higher 0-1 loss. When stratified by hospital and population composition, these models demonstrated heterogeneous algorithmic bias in different contexts and populations.

CONCLUSIONS

Caution must be taken when interpreting fairness measures' face value. A higher FNR or FPR could potentially reflect missed opportunities or wasted resources, but these measures could also reflect health care use patterns and gaps in care. Simply relying on the statistical notions of bias could obscure or underplay the causes of health disparity. The imperfect health data, analytic frameworks, and the underlying health systems must be carefully considered. Fairness measures can serve as a useful routine assessment to detect disparate model performances but are insufficient to inform mechanisms or policy changes. However, such an assessment is an important first step toward data-driven improvement to address existing health disparities.

Comparing approaches to identify live births using the Transformed Medicaid Statistical Information System

OBJECTIVE

To evaluate the performance of different approaches for identifying live births using Transformed Medicaid Statistical Information System Analytic Files (TAF).

DATA SOURCES

The primary data source for this study were TAF inpatient (IP), other services (OT), and demographic and eligibility files. These data contain administrative claims for Medicaid enrollees in all 50 states and the District of Columbia from January 1, 2018 to December 31, 2018.

STUDY DESIGN

We compared five approaches for identifying live birth counts obtained from the TAF IP and OT data with the Centers for Disease Control and Prevention (CDC) Natality data-the gold standard for birth counts at the state level.

DATA COLLECTION/EXTRACTION METHODS

The five approaches used varying combinations of diagnosis and procedure, revenue, and place of service codes to identify live births. Approaches 1 and 2 follow guidance developed by the Centers for Medicare and Medicaid Services (CMS). Approaches 3 and 4 build on the approaches developed by CMS by including all inpatient hospital claims in the OT file and excluding codes related to delivery services for infants, respectively. Approach 5 applied Approach 4 to only the IP file.

PRINCIPAL FINDINGS

Approach 4, which included all inpatient hospital claims in the OT file and excluded codes related to infants to identify deliveries, achieved the best match of birth counts relative to CDC birth record data, identifying 1,656,794 live births-a national overcount of 3.6%. Approaches 1 and 3 resulted in larger overcounts of births (20.5% and 4.5%), while Approaches 2 and 5 resulted in undercounts of births (-3.4% and -6.8%).

CONCLUSIONS

Including claims from both the IP and OT files, and excluding codes unrelated to the delivery episode and those specific to services rendered to infants improves accuracy of live birth identification in the TAF data.

Incidence and Clinical Risk Factors of Post-Operative Complications following Primary Total Hip Arthroplasty: A 10-Year Population-Based Cohort Study

Background: Total hip arthroplasty (THA) has become a growing treatment procedure for debilitating hip pathologies. Patients experienced post-operative complications and revision surgeries according to large THA registries. To fully understand the short-term and long-term post-operative outcomes following THA, the purpose of this study is to examine the incidence of post-operative complications following primary THA and to examine how this trend has changed over 10 years within community hospitals in the US using large databases. Methods: This study queried the State Inpatient Database (SID) for primary THA between 2006 and 2015. Individual patients were followed forward in time until the first instance of a post-operative complication. The multivariable logistic regression analyses were computed to examine which post-operative complications were independent predictors of pre-operative comorbidities. Results: Median age of patients was 67 years, and 56% of patients were female. Females with avascular necrosis (AVN) as an indication for THA had a 27% higher risk of complication. Females with osteoarthritis (OA) as an indication for THA had a 6% higher risk of complication. Post-operative complications occurred with higher frequencies in the first two months of THA and the highest risks of THA complications within the first 6 months. Conclusion: The most common indication is OA in elders with primary THA. Females and those of black ethnicity showed the greatest risks of THA complications. Data from our large study can be used to understand post-operative complications and readmissions after THA. Our study also provides data on risk factors associated with these complications.

Association of Do-Not-Resuscitate orders and in-hospital mortality among patients undergoing cranial neurosurgery

BACKGROUND

Previous studies identified pre-existing DNR orders as a predictor of mortality after surgery. We sought to evaluate mortality of patients receiving cranial neurosurgery with DNR orders placed at the time of, or within 24 h of admission.

METHODS

We performed a retrospective cohort study using the California State Inpatient Database, January 2018 to December 2020. We used International Classification of Diseases, 10th Revision (ICD-10) codes to identify emergent hospitalizations with principal diagnosis of brain injury, including traumatic brain injury [TBI], ischemic stroke [IS], intracerebral hemorrhage [ICH], subarachnoid hemorrhage [SAH], or malignant brain tumor [mBT]. We used procedure and Diagnosis Related Group codes to identify cranial neurosurgery. Patients with DNR were one-to-one matched to non-DNR controls based on diagnosis (exact matching), age, sex, Elixhauser comorbidity index, and organ failure (coarsened matching). The primary outcome was inpatient mortality.

RESULTS

In California, 30,384 patients underwent cranial neurosurgery, 2018-2020 (n = 3,112, 10% DNR). DNR patients were older, more often female, more often White, with greater comorbidity and organ system dysfunction. There were 2,505 patients with DNR orders 1:1 matched to controls. Patients with DNR had greater inpatient mortality (56% vs. 23%, p < 0.001; Hazard Ratio 3.11, 95% CI 2.50-3.86), received tracheostomy (Odds Ratio [OR] 0.37, 95% CI 0.24-0.57) and gastrostomy less (OR 0.48, 95% CI 0.39-0.58) compared to controls. Multivariable analysis of the unmatched cohort demonstrated similar results.

CONCLUSION

Patients undergoing cranial neurosurgery with early or pre-existing DNR have high inpatient mortality compared to clinically similar non-DNR patients; 1 in 2 died during their hospitalization.

Age- and Sex-Specific Trends in Medical Complications After Acute Ischemic Stroke in the United States

BACKGROUND AND OBJECTIVES

To test the hypothesis that the age and sex-specific prevalence of infectious (pneumonia, sepsis, and urinary tract infection [UTI]) and noninfectious (deep venous thrombosis [DVT], pulmonary embolism [PE], acute renal failure [ARF], acute myocardial infarction [AMI], and gastrointestinal bleeding [GIB]) complications increased after acute ischemic stroke (AIS) hospitalization in the United States from 2007 to 2019.

METHODS

We conducted a serial cross-sectional study using the 2007-2019 National Inpatient Sample. Primary AIS admissions in adults (aged 18 years or older) with and without complications were identified using International Classification of Diseases codes. We quantified the age/sex-specific prevalence of complications and used negative binomial regression models to evaluate trends over time.

RESULTS

Of 5,751,601 weighted admissions, 51.4% were women. 25.1% had at least 1 complication. UTI (11.8%), ARF (10.1%), pneumonia (3.2%), and AMI (2.5%) were the most common complications, while sepsis (1.7%), GIB (1.1%), DVT (1.2%), and PE (0.5%) were the least prevalent. Marked disparity in complication risk existed by age/sex (UTI: men 18-39 years 2.1%; women 80 years or older 22.5%). Prevalence of UTI (12.9%-9.7%) and pneumonia (3.8%-2.7%) declined, but that of ARF increased by ≈3-fold (4.8%-14%) over the period 2007-2019 (all p < 0.001). AMI (1.9%-3.1%), DVT (1.0%-1.4%), and PE (0.3%-0.8%) prevalence also increased (p < 0.001), but that of sepsis and GIB remained unchanged over time. After multivariable adjustment, risk of all complications increased with increasing NIH Stroke Scale (pneumonia: prevalence rate ratio [PRR] 1.03, 95% CI 1.03-1.04, for each unit increase), but IV thrombolysis was associated with a reduced risk of all complications (pneumonia: PRR 0.80, 85% CI 0.73-0.88; AMI: PRR 0.85, 95% CI 0.78-0.92; and DVT PRR 0.87, 95% CI 0.78-0.98). Mechanical thrombectomy was associated with a reduced risk of UTI, sepsis, and ARF, but DVT and PE were more prevalent in MT hospitalizations compared with those without. All complications except UTI were associated with an increased risk of in-hospital mortality (sepsis: PRR 1.97, 95% CI 1.78-2.19).

DISCUSSION

Infectious complications declined, but noninfectious complications increased after AIS admissions in the United States in the last decade. Utilization of IV thrombolysis is associated with a reduced risk of all complications.

Trends in gastrointestinal disease hospitalizations and outcomes during the first year of the coronavirus pandemic

BACKGROUND

The impact of the coronavirus on hospitalizations for gastrointestinal (GI) disease, particularly at a population level is understudied.

AIM

To investigate trends in hospitalizations, inpatient endoscopy resource utilization, and outcomes during the first year of the coronavirus pandemic and subsequent lockdowns.

METHODS

Using the California State Inpatient Database for 2018-2020, we explored year-to-year and 2020 month-to-month trends in hospitalizations, length of stay, and inpatient mortality (all-cause & viral pneumonia-specific) for common inpatient GI diagnoses including acute pancreatitis, diverticulitis, cholelithiasis, non-infectious gastroenteritis, upper and lower GI bleeding (LGIB), Clostridium difficile, viral gastroenteritis, inflammatory bowel disease, and acute cholangitis.

RESULTS

Disease-specific hospitalizations decreased for all included conditions except nonvariceal upper GI bleeding (NVUGIB), LGIB, and ulcerative colitis (UC) (ptrend < 0.0001). All-cause inpatient mortality was higher in 2020 vs 2019, for acute pancreatitis (P = 0.029), diverticulitis (P = 0.04), NVUGIB (P = 0.003), and Crohn’s disease (P = 0.004). In 2020, hospitalization rates were lowest in April, November, and December. There was no significant corresponding increase in inpatient mortality except in UC (ptrend = 0.048). Viral pneumonia and viral pneumonia complicated by respiratory failure increased (P < 0.001) among GI hospitalizations. Endoscopy utilization within 24 h of admission was unchanged for GI emergencies except NVUGIB (P < 0.001).

CONCLUSION

Our findings suggest that hospitalization rates for common GI conditions significantly declined in California during the COVID pandemic, particularly in April, November and December 2020. All-cause mortality was significantly higher among acute pancreatitis, diverticulitis, NVUGIB, and Crohn’s disease hospitalizations. Emergency endoscopy rates were mostly comparable between 2020 and 2019.

A Practical Guide to Use of Publicly Available Data Sets for Observational Research in Interventional Radiology

Observational data research studying access, utilization, cost, and outcomes of image-guided interventions using publicly available "big data" sets is growing in the interventional radiology (IR) literature. Publicly available data sets offer insight into real-world care and represent an important pillar of IR research moving forward. They offer insights into how IR procedures are being used nationally and whether they are working as intended. On the other hand, large data sources are aggregated using complex sampling frames, and their strengths and weaknesses only become apparent after extensive use. Unintentional misuse of large data sets can result in misleading or sometimes erroneous conclusions. This review introduces the most commonly used databases relevant to IR research, highlights their strengths and limitations, and provides recommendations for use. In addition, it summarizes methodologic best practices pertinent to all data sets for planning and executing scientifically rigorous and clinically relevant observational research.

National recurrence of pancreatitis and readmissions after biliary pancreatitis

BACKGROUND

National and international guidelines support early cholecystectomy after mild gallstone pancreatitis but a recent nationwide study suggested these recommendations are not universally followed. Our study sought to quantify the national utilization of same hospitalization cholecystectomy versus non-operative management (NOM) and its association with pancreatitis recurrence, readmissions, and costs after mild gallstone pancreatitis (GP).

METHODS

Adult patients admitted with mild GP were identified from the Nationwide Readmission Database 2010-2015. Primary outcomes included the rate of cholecystectomy during the index admission as well as pancreatitis recurrence and readmission at 30 and 180 days (30d, 180d) comparing NOM to same hospitalization cholecystectomy. Mortality upon readmission, total length of stay (LOS), and total costs (combined index-readmission hospital costs) were also explored. Cox proportional hazards regression and generalized linear models controlled for patient/hospital confounders.

RESULTS

Among the 65,067 patients identified, 30% underwent cholecystectomy. The NOM cohort was older (58 vs. 50 years), had more comorbidities (Charlson index > 2, 23.5% vs. 11.5%), fewer female patients (56.7% vs. 67%) and less discharge-to-home (84.9% vs. 94.4%) (all p < 0.001). NOM was associated with increase in recurrence and unplanned readmissions at 30d [Hazard Ratio 3.53 (95% CI 2.92-4.27), 2.41 (2.11-2.74), respectively], and 180d [4.27 (3.65-4.98), 2.78 (2.54-3.04), respectively], as well as increased mortality during 180d readmission 1.88 (1.06-3.35). This approach was also associated with significant increase in LOS [predicted mean difference 2.79 days (95% CI 2.46-3.12)] and total costs [$2507.89 ($1714.4-$3301.4)].

CONCLUSIONS

In the USA, most patients presenting with mild GP do not undergo same hospitalization cholecystectomy. This strategy results in higher recurrent pancreatitis, mortality during readmission, and an additional $4.85 M/year in hospital costs nationwide. These data support same hospitalization cholecystectomy as the gold standard for mild GP.