This cross-sectional study examined data collected for patients aged 18 and older who underwent inpatient or outpatient OSA surgical procedures as defined by International Classification of Diseases, 9th Revision, Clinical Modification (ICD-9-CM) diagnosis and procedure codes.
Specifically, patients were selected if they had an OSA diagnosis code (327.23, 780.50, 780.51, 780.53, 780.57, 780.59, or 786.03), did not have a diagnosis code for head and neck neoplasm, and underwent a palate or hypopharyngeal OSA surgery. Palate surgery procedures were defined using ICD-9-CM codes 27.64, 27.69, 27.72, 27.73, or 29.4. Hypopharyngeal procedures were defined as tongue radiofrequency or midline glossectomy (25.1, 25.2, 25.59, 25.94, or 25.99); lingual tonsillectomy (28.5); genioglossus advancement, genioplasty, or tongue stabilization (76.63, 76.64, 76.67, or 76.68); and hyoid suspension (83.02). Maxillomandibular advancement (76.43, 76.46, 76.61, 76.62, 76.65, and/or 76.66) was included as a separate category.
None of the data used in this research contained personal identifying information, and the study was exempt from UCSF institutional review.
No single national database captures both inpatient and outpatient surgical procedures for the entire study period. Therefore, this study required the combination of a national database for inpatient procedures and separate state-level databases for inpatient and outpatient procedures.
Inpatient OSA surgeries were examined using the Nationwide Inpatient Sample (NIS) for 2000, 2004, and 2006. The NIS contains patient demographics, diagnosis and procedure codes, insurance type, and facility characteristics on all inpatient stays from a 20% stratified sample of hospitals from 28 (2000), 37 (2004), or 38 (2006) states. The NIS does not include overnight admissions after procedures performed in acute care hospitals that are coded as observation status (as opposed to inpatient status, with its greater intensity of monitoring); these observation status admissions are not included in this study. Inverse-probability-of-sampling weights are provided with the NIS data, enabling users to estimated total nationwide inpatient procedure volumes.
Outpatient procedures for 2006 were identified using data from State Ambulatory Surgery Database (SASD) and State Inpatient Database (SID) files for California, New York, North Carolina, and Wisconsin. These states were selected in order to gain a wide geographic distribution and because these states’ data capture encounters from both hospital-based and freestanding ambulatory surgery centers in the former database. Data for 2000 and 2004 were not used because the relevant databases either do not exist or have substantial missing data. No weighting is required because both databases capture all outpatient (ambulatory surgery) encounters or inpatient admissions, respectively, from facilities in each state.
The 2004 and 2006 NIS contain billed hospital charges and hospital-specific cost-to-charge ratios for the majority of OSA procedures, enabling an evaluation of inpatient procedure costs from the perspective of the hospital. Cost-to-charge ratios were used to convert billed charges to costs and then adjusted using the medical component of the consumer price index.4
Outcomes of interest for this study were procedure volumes, whether the patient underwent palate or hypopharyngeal surgery, and whether or not the procedure was performed in the inpatient or outpatient setting. We also examined (in national data for 2004 and 2006 only) costs for inpatient surgery admissions.
We examined a number of independent patient and hospital variables that were chosen based on their potential associations with the outcome measures. The key independent variables were selected as a subset of all variables within the database.
Patient variables of interest included age, gender, race/ethnicity, insurance type, type of county of residence according to the Department of Agriculture Urban Influence Code categories,5
and median annual household income for patient zip code. Medical comorbidities were identified using the Elixhauser method;6
categories were defined as no (0 conditions), low (1–2 conditions), or moderate to high (≥3 conditions) comorbidity, similar to previous analyses using the NIS.7
Hospital level predictors included location/teaching status (defined as urban teaching, urban non-teaching, or rural) and region: Northeast (including New York), Midwest (including Wisconsin), South (including North Carolina), and West (including California).
The statistical analyses were completed using a combination of databases (). All statistical analyses of the NIS data were conducted using methods for weighted complex surveys. Inpatient procedures were analyzed using the NIS data for 2000, 2004, and 2006. The following surgical volumes were estimated using the NIS weights, as described above: total procedures, specific procedures, and procedures according to the key independent variables. Chi-square tests for trend compared the number of procedures and the distribution of procedures among categories of a specific variable (e.g., age groups) across the time period. Multiple logistic regression examined the association between the performance of isolated soft palate surgery (vs. hypopharyngeal procedure with or without palate surgery) and the key independent variables (simultaneous adjustment). Interaction terms with year were included to determine whether the expected values of the outcomes changed over time.
Databases used in statistical analyses, according to outcome measure
Costs for inpatient procedure admissions were also calculated for 2004 and 2006, with reporting of means and 95% confidence intervals (95% CI). Hospital-specific cost-to-charge ratios contained within the NIS were used to convert billed charges to costs. To compare costs across years, costs for 2006 were divided by 1.12, adjusting for medical price inflation according to the Bureau of Labor Statistics’ medical component of the Consumer Price Index (derived from reimbursements).4
Mean overall costs (in 2004 dollars) and those associated with isolated palate and hypopharyngeal (with or without palate surgery) were calculated, then compared across years and procedure categories. Because of their right-skewed distribution, costs were log-transformed to meet the assumptions of the multiple linear regression, which was then used to examine the independent association between costs and type of surgery (isolated palate vs. hypopharyngeal +/− palate surgery), simultaneously adjusting for the following potential confounders: age group, gender, race/ethnicity, medical comorbidity, median household income for zip code, primary payment source, and hospital location/teaching status.
Outpatient procedures were incorporated using 2006 state-level data from the SASD and SID files for California, New York, North Carolina, and Wisconsin. Multiple steps were required to use the state-level data to estimate total numbers of inpatient plus outpatient procedures.
First, the four state-level databases were combined. Procedures of interest were identified using the same combinations of diagnosis and procedure codes as in the NIS database. Multiple logistic regression was then used to estimate the probability that each procedure was performed in an inpatient rather than an outpatient setting, as a function of state, procedure type (palate vs. other), primary payment source, and patient age, gender, and county of residence, classified as in the NIS data. Race/ethnicity was not used in this procedure because it was missing for a large proportion of the outpatient procedures.
We then used the coefficients from these logistic models, in conjunction with the same covariates for each procedure in the NIS, to estimate the probability that each of those NIS procedures, considered as part of the universe of all such procedures, had been performed in an inpatient setting. In turn, we used those probabilities to rescale the original NIS weights and estimate total procedure volume.
To see how this works, suppose that the NIS weight for a selected observation was 5, meaning that the NIS observation represents an estimated total of 5 inpatient procedures, reflecting the NIS 20% systematic sampling. Suppose also that the estimated probability that this particular procedure was performed in an inpatient setting is 25%. Accordingly, the final weight for this observation would be calculated as 5/0.25 = 20. Thus from this one NIS observation we would project a total of 20 procedures, including 5 inpatient and 15 outpatient.
Next, we repeated the analyses using the new weights, now reflecting combined inpatient and outpatient procedures for 2006. Thus in this analysis, we estimated combined procedure volumes and the independent associations between performance of any inpatient and outpatient isolated soft palate surgery (vs. hypopharyngeal procedure with or without palate surgery) and the key independent variables.
Statistical analyses were performed using the SUDAAN (Research Triangle Institute, Research Triangle Park, NC) statistical software Version 10.0 and SAS Version 9.2 (SAS Institute, Cary, NC). P-values < 0.05 were considered statistically significant.