Mortality for Time-Sensitive Conditions at Urban vs Rural Hospitals During the COVID-19 Pandemic

Key Points Question How did patient outcomes for non–COVID-19 time-sensitive conditions change during the pandemic? Findings In this cohort study of 3813 US hospitals, including 18 601 925 hospitalizations, odds of in-hospital mortality increased significantly for non–COVID-19 sepsis and pneumonia during the pandemic (March 8, 2020, to December 31, 2021) compared with before the pandemic (January 1, 2017, to March 7, 2020) at both urban and rural hospitals, but the increase was greater at rural hospitals. In-hospital mortality significantly increased for acute myocardial infarction and gastrointestinal bleeding at urban hospitals and for hip fracture at rural hospitals. Meaning In this study, patient outcomes for time-sensitive conditions were worse during the pandemic than before, with different magnitudes of change at urban vs rural hospitals, suggesting that strategies tailored to the different needs of these hospitals may reduce mortality during future public health crises.

The analysis used an interrupted time-series logistic regression model specified as follows: logit(diedit) = β0 + β1Rural + β2t + β3t*Rural + β4m + β5X + β6X*Rural where diedit is the probability of death for person i at time t.The model included a term for rural status of the hospital, defined as being in a ZIP Code eligible for funding from the Federal Office of Rural Health Policy. 2 It included a continuous measure of time t with values 1 through 60, defined as the month of discharge where t=1 for discharges in January 2017 and where the value increases by 1 per month.This accounts for the underlying pre-period trend removing seasonality effects.We interacted t with Rural, allowing for rural/urban-specific pre-period slopes.The model also included β4m, a fixed effect for calendar month (12 months), which accounts for seasonality effects that are assumed constant in the preand peri-pandemic periods.The model included a dichotomous indicator X (0/1), with 0 for the discharge in the prepandemic period (January 1, 2017-March 7, 2020) and 1 for the peri-pandemic period of 2020 (March 8-December 31, 2020).The main effect of interest, β5, corresponds to the level shift in the log odds of the death rate associated with the COVID-19 pandemic, comparing inpatient stays in the peripandemic period of 2020 with what would have been expected based on mortality rates in the prepandemic period.We interacted X with Rural to obtain separate effects for urban hospitals (parameterized by β5) and rural hospitals (parameterized by β5 + β6).Standard errors were clustered on hospital ID.The models were weighted using entropy weights 3 that were calculated separately for stays for each condition (AMI, GI hemorrhage, hip fracture, pneumonia, sepsis, stroke), aligning age, sex, and the comorbidity score, calculated using the Elixhauser Comorbidity Software Refined for ICD-10, 4 among stays in the prepandemic period with stays in the 2020 pandemic period.
To obtain separate effects during the 2020 pandemic period by Covid-19 burden in the hospital's community, we modified the model so that X was replaced by the COVID-19 burden variable (0/1), with value 1 for each of the four COVID burden levels (i.e., low, moderate, substantial, or high), and 0 for stays in the prepandemic period.We ran four models estimating the effects of the pandemic for each burden level separately, using burden-specific weights.Separate entropy weights were calculated aligning age, sex, and the comorbidity score for stays in the prepandemic period with peri-pandemic stays for each level of burden.
To obtain separate effects during the 2020-2021 peri-pandemic period by month, we modified the model so that X was replaced by the interaction of month m and the peri-pandemic indicator X. Monthly effects were estimated using quarterly data.There were seven quarters from March 2020 to December 2021 during the pandemic.Separate entropy weights were calculated comparing stays in each peripandemic quarter of 2020 (3 quarters) and 2021 (all 4 quarters) with those in the prepandemic period.
We then ran quarterly models generating effect estimates for each month of that quarter separately, using quarter-specific weights.The entropy and interrupted time-series models for each quarter of 2021 excluded stays in the prepandemic period if they were in states that did not contribute to the Healthcare Cost and Utilization Project (HCUP) State Inpatient Databases (SID) in that quarter (see eTable 1).Note: Values in bold are statistically significant (p<0.05).Odds ratio estimates are from interrupted time-series logistic regression models.Separate models were estimated for each month and included the following independent variables: rural status (based on hospital location), peri-period (discharges after March 8, 2020), and an interaction between peripandemic and rural.All models were weighted for pre/peri differences in demographic and clinical characteristics using entropy weights calculated separately for stays for each condition, aligning age, sex, and Elixhauser comorbidity index score for mortality.Standard errors were clustered on hospital ID.

eFigure 1 .
Odds of In-Hospital Mortality Among Non-Covid-19 Stays for GI Hemorrhage, Hip Fracture, and Stroke During March 8-December 31, 2020, Overall and by Covid-19 Burden in the Hospital's Community, Relative to Prepandemic Stays.Notes: CI, confidence interval; OR, odds ratio.Because not all states had data in 2021, these results are limited to the 2020 peri-pandemic period (March 8-December 31, 2020).

eTable 1 .
Number of States Available by Year and Quarter S72101C Unspecified trochanteric fracture of right femur, initial encounter for open fracture type IIIA, IIIB, or IIIC S72102A Unspecified trochanteric fracture of left femur, initial encounter for closed fracture S72102B Unspecified trochanteric fracture of left femur, initial encounter for open fracture type I or II Characteristics of Inpatient Stays in the Prepandemic Period (January 1, 2017-March 7, 2020) and Inpatient Stays Without a Covid-19 Diagnosis in the Peri-pandemic Period (March 8, 2020-December 31, 2021) Full Regression Models Comparing In-Hospital Mortality Among Stays During March 8-December 31, 2020, Overall and by Covid-19 Burden in the Hospital's Community, Relative to Prepandemic Stays Full Regression Models Comparing In-Hospital Mortality Among Non-Covid-19 Stays in 2020 and 2021, by Month, Relative to Prepandemic stays S72011A Unspecified intracapsular fracture of right femur, initial encounter for closed fracture S72011B Unspecified intracapsular fracture of right femur, initial encounter for open fracture type I or II S72011C Unspecified intracapsular fracture of right femur, initial encounter for open fracture type IIIA, IIIB, or IIIC S72012A Unspecified intracapsular fracture of left femur, initial encounter for closed fracture S72023B Displaced fracture of epiphysis (separation) (upper) of unspecified femur, initial encounter for open fracture type I or II S72023C Displaced fracture of epiphysis (separation) (upper) of unspecified femur, initial encounter for open fracture type IIIA, IIIB, or IIIC S72024A Nondisplaced fracture of epiphysis (separation) (upper) of right femur, initial encounter S72063A Displaced articular fracture of head of unspecified femur, initial encounter for closed fracture S72063B Displaced articular fracture of head of unspecified femur, initial encounter for open fracture type I or II S72063C Displaced articular fracture of head of unspecified femur, initial encounter for open fracture type IIIA, IIIB, or IIIC S72064A Nondisplaced articular fracture of head of right femur, initial encounter for closed fracture S72064B Nondisplaced articular fracture of head of right femur, initial encounter for open fracture type I or II S72064C Nondisplaced articular fracture of head of right femur, initial encounter for open fracture type IIIA, IIIB, or IIIC S72065A Nondisplaced articular fracture of head of left femur, initial encounter for closed fracture S72065B Nondisplaced articular fracture of head of left femur, initial encounter for open fracture type I or II S72065C Nondisplaced articular fracture of head of left femur, initial encounter for open fracture type IIIA, IIIB, or IIIC S72066A Nondisplaced articular fracture of head of unspecified femur, initial encounter for closed fracture S72066B Nondisplaced articular fracture of head of unspecified femur, initial encounter for open fracture type I or II S72066C Nondisplaced articular fracture of head of unspecified femur, initial encounter for open fracture type IIIA, IIIB, or IIIC S72091A Other fracture of head and neck of right femur, initial encounter for closed fracture S72091B Other fracture of head and neck of right femur, initial encounter for open fracture type I or II S72091C Other fracture of head and neck of right femur, initial encounter for open fracture type IIIA, IIIB, or IIIC S72092A Other fracture of head and neck of left femur, initial encounter for closed fracture S72092B Other fracture of head and neck of left femur, initial encounter for open fracture type I or II S72099C Other fracture of head and neck of unspecified femur, initial encounter for open fracture type IIIA, IIIB, or IIIC S72101A Unspecified trochanteric fracture of right femur, initial encounter for closed fracture S72101B Unspecified trochanteric fracture of right femur, initial encounter for open fracture type I or II ICD-10-CM code Description a Self-pay/no charge: includes self-pay, no charge, charity, and no expected payment.eTable4.Abbreviations: AMI, acute myocardial infarction; CI, confident interval; GI, gastrointestinal; OR, odds ratio Note: Values in bold are statistically significant (p<0.05).Odds ratio estimates are from interrupted time-series logistic regression models.Each model included the following independent variables: rural status (based on hospital location), month of discharge (continuous 1-60 starting in January 2017), calendar month (values 1 through 12), peri-pandemic period (discharges after March 8, 2020), and an interaction between peri-pandemic and rural.All models were weighted for pre/peri differences in demographic and clinical characteristics using entropy weights calculated separately for stays for each condition, aligning age, sex, and Elixhauser comorbidity index score for mortality.Standard errors were clustered on hospital ID.eTable 5.