Nationwide Inpatient Sample and National Surgical Quality Improvement Program give different results in hip fracture studies

Abstract

Background: National databases are being used with increasing frequency to conduct orthopaedic research. However, there are important differences in these databases, which could result in different answers to similar questions; this important potential limitation pertaining to database research in orthopaedic surgery has not been adequately explored.

Questions/purposes: The purpose of this study was to explore the interdatabase reliability of two commonly used national databases, the Nationwide Inpatient Sample (NIS) and the National Surgical Quality Improvement Program (NSQIP), in terms of (1) demographics; (2) comorbidities; and (3) adverse events. In addition, using the NSQIP database, we identified (4) adverse events that had a higher prevalence after rather than before discharge, which has important implications for interpretation of studies conducted in the NIS.

Methods: A retrospective cohort study of patients undergoing operative stabilization of transcervical and intertrochanteric hip fractures during 2009 to 2011 was performed in the NIS and NSQIP. Totals of 122,712 and 5021 patients were included from the NIS and NSQIP, respectively. Age, sex, fracture type, and lengths of stay were compared. Comorbidities common to both databases were compared in terms of more or less than twofold difference between the two databases. Similar comparisons were made for adverse events. Finally, adverse events that had a greater postdischarge prevalence were identified from the NSQIP database. Tests for statistical difference were thought to be of little value given the large sample size and the resulting fact that statistical differences would have been identified even for small, clinically inconsequential differences resulting from the associated high power. Because it is of greater clinical importance to focus on the magnitude of differences, the databases were compared by absolute differences.

Results: Demographics and hospital lengths of stay were not different between the two databases. In terms of comorbidities, the prevalences of nonmorbid obesity, coagulopathy, and anemia in found in the NSQIP were more than twice those in the NIS; the prevalence of peripheral vascular disease in the NIS was more than twice that in the NSQIP. Four other comorbidities had prevalences that were not different between the two databases. In terms of inpatient adverse events, the frequencies of acute kidney injury and urinary tract infection in the NIS were more than twice those in the NSQIP. Ten other inpatient adverse events had frequencies that were not different between the two databases. Because it does not collect data after patient discharge, it can be implied from the NSQIP data that the NIS does not capture more than ½ of the deaths and surgical site infections occurring during the first 30 postoperative days.

Conclusions: This study shows that two databases commonly used in orthopaedic research can identify similar populations of operative patients but may generate very different results for specific commonly studied comorbidities and adverse events. The NSQIP identified higher rates of morbid obesity, coagulopathy, and anemia. The NIS identified higher rates of peripheral vascular disease, acute kidney injury, and urinary tract infection.

Level of evidence: Level II, prognostic study. See the Instructions for Authors for a complete description of levels of evidence.

Fig. 1

The numbers of orthopaedic surgery publications using the NIS and NSQIP per year since 2000 based on a PubMed search of the 41 orthopaedic journals identified by Moverley et al. [12] for the terms “Nationwide Inpatient Sample” or “National Surgical Quality Improvement Program” on November 19, 2013 are shown. Only original research studies were included. In total, there were 72 studies using the NIS and 12 using the NSQIP.

Fig. 2A–B

The distributions of patient age for the (A) NIS and (B) NSQIP are shown. The 25th (74 and 75 years), 50th (82 and 83 years), and 75th (88 and 89 years) percentiles were within 1 year of each other. To make patients less identifiable, the NSQIP reports all patients older than 90 years as 90 years; to make data comparable, we treated patients in the NIS who were older than 90 years as 90 years. The regularly spaced spikes in the distributions are the result of some states’ requirements that age be rounded to make data less identifiable.

Fig. 3A–B

Survival curves for length of stay for the (A) NIS and (B) NSQIP are shown. The numbers of days after the procedure at which 75% (4 and 4 days), 50% (5 and 5 days), and 25% (7 and 7 days) of patients remained in the hospital were identical.

Fig. 4

For comorbidities shown in the upper 1/3, the rate documented in the NSQIP was more than twice that in the NIS. For comorbidities shown in the lower 1/3, the rate documented in the NIS was more than twice that in the NSQIP. In the middle 1/3, the rates documented in the two databases were within a twofold difference of each other.

Fig. 5

For events shown in the lower ½, the rate documented in the NIS was more than twice the rate documented in the NSQIP before discharge. For events shown in the upper ½, the rate documented in the NIS and the rate documented in the NSQIP before discharge were within a twofold difference of each other. For the NSQIP, the percent of patients with adverse events after discharge can be added to the percent of patients with adverse events before discharge to get the total rate of adverse events during the first 30 postoperative days.