We conducted a retrospective, serial cross-sectional analysis on a 20% sample of all delivery hospitalization discharges in the United States between 2005 and 2014 using the Healthcare Cost and Utilization Project’s National Inpatient Sample, compiled by the Agency for Healthcare Research and Quality. The National Inpatient Sample is the largest nationally-representative sample of hospital discharges in the United States.13 The dataset contains clinical and non-clinical data for each hospitalization, including diagnostic and procedure codes, patient demographic characteristics, and expected payment source. Deliveries were identified by ICD-9 codes using previously described methods.14 (link) Because the analysis was of de-identified national data, our study was deemed exempt from review by the Institutional Review Board at the University of Michigan Medical School.
We examined the prevalence of eight common, chronic conditions known to impact obstetric morbidity and mortality: chronic respiratory disease, chronic hypertension, substance use disorders, pre-existing diabetes, chronic heart disease, chronic renal disease, chronic liver disease, and HIV. We focused our analyses on pre-existing conditions rather than pregnancy-related conditions. Chronic conditions were defined using the ICD-9 codes listed inTable 1 , which were chosen for inclusion based on review of literature13 ,14 (link) and author consensus. Sub-group analysis of ICD-9-CM codes included in the variable for chronic respiratory disease revealed that this variable was comprised almost entirely of observations with diagnosis codes for asthma (493.x) and will be referred to as asthma for the remainder of this report.
We controlled for several covariates in our analyses, including age, rural vs. urban residence, median household income quartile for the patient’s ZIP Code, and primary payer. Given that a number of hospitals and HCUP partners do not report data on race or ethnicity, we were unable to include these co-variates.13 Location of residence was defined as rural or urban using the National Center for Health Statistics Classification and Urban Influence Codes.15 Payment sources were grouped into public insurance (Medicaid and Medicare), private insurance, and uninsured or self-pay. Given that fewer than 0.6% of the delivery hospitalizations were funded by Medicare, public sources are referred to as Medicaid throughout the study. The number of observations with missing values was approximately 2% of all delivery hospitalizations, which was considered sufficient for analysis.
We used multivariable logistic regression models with predictive margins to obtain estimates of disease-specific prevalence and to estimate the rates at which any one and multiple chronic conditions were identified per 1,000 delivery hospitalizations. Data was pooled into two-year periods to increase the precision of our estimates. We estimated disease-specific prevalence by key socio-economic predictors for the four most prevalent conditions by interacting rural vs. urban residence, income, and payer with time in adjusted multivariable logistic regression models. Predictive margins were used in all sub-group analyses to generate adjusted prevalence estimates. We examined differences in prevalence for each condition over time by subgroup. We compared changes in prevalence over time across subgroups using a difference-in-differences framework.
We utilized National Inpatient Sample trend weights to allow for comparisons across years. Results are weighted to allow for nationally-representative inferences unless otherwise noted. Full details about sampling and weighting procedures are available at the Healthcare Cost and Utilization Project website.13 Two-sided P values <.05 were considered statistically significant. All analyses were conducted using STATA version 14.2 (StataCorp, College Station, TX).
We examined the prevalence of eight common, chronic conditions known to impact obstetric morbidity and mortality: chronic respiratory disease, chronic hypertension, substance use disorders, pre-existing diabetes, chronic heart disease, chronic renal disease, chronic liver disease, and HIV. We focused our analyses on pre-existing conditions rather than pregnancy-related conditions. Chronic conditions were defined using the ICD-9 codes listed in
We controlled for several covariates in our analyses, including age, rural vs. urban residence, median household income quartile for the patient’s ZIP Code, and primary payer. Given that a number of hospitals and HCUP partners do not report data on race or ethnicity, we were unable to include these co-variates.13 Location of residence was defined as rural or urban using the National Center for Health Statistics Classification and Urban Influence Codes.15 Payment sources were grouped into public insurance (Medicaid and Medicare), private insurance, and uninsured or self-pay. Given that fewer than 0.6% of the delivery hospitalizations were funded by Medicare, public sources are referred to as Medicaid throughout the study. The number of observations with missing values was approximately 2% of all delivery hospitalizations, which was considered sufficient for analysis.
We used multivariable logistic regression models with predictive margins to obtain estimates of disease-specific prevalence and to estimate the rates at which any one and multiple chronic conditions were identified per 1,000 delivery hospitalizations. Data was pooled into two-year periods to increase the precision of our estimates. We estimated disease-specific prevalence by key socio-economic predictors for the four most prevalent conditions by interacting rural vs. urban residence, income, and payer with time in adjusted multivariable logistic regression models. Predictive margins were used in all sub-group analyses to generate adjusted prevalence estimates. We examined differences in prevalence for each condition over time by subgroup. We compared changes in prevalence over time across subgroups using a difference-in-differences framework.
We utilized National Inpatient Sample trend weights to allow for comparisons across years. Results are weighted to allow for nationally-representative inferences unless otherwise noted. Full details about sampling and weighting procedures are available at the Healthcare Cost and Utilization Project website.13 Two-sided P values <.05 were considered statistically significant. All analyses were conducted using STATA version 14.2 (StataCorp, College Station, TX).
