by Devon J. Ryan, BA, JT Stranix, MD (@jtstranix)
Empirical studies using national datasets are increasingly prevalent in the plastic surgery literature. A working knowledge of these databases is therefore essential to performing and analyzing these investigations. This review provides an introduction to the two primary databases currently available and highlights their use in the plastic surgery literature.
The National Inpatient Sample (NIS) is a publicly available database maintained by the Agency for Healthcare Research & Quality (AHRQ), a government agency, as part of the Healthcare Cost & Utilization Project (HCUP).1 Sampling 20% of US hospital admissions, it collects data from over 7 million hospitalizations annually and can be weighted to obtain estimates of approximately 36 million hospital stays per year. Information available includes sociodemographic information, diagnosis/treatment codes, complications, cost, and hospital characteristics. Data can be purchased by year, with 1988-2012 currently available.
The National Surgical Quality Improvement Program (NSQIP) database is also publicly available, arising from an American College of Surgeons (ACS) quality improvement initiative.2,3 Unlike the NIS, however, hospitals included in NSQIP must volunteer to join the program – producing a potentially biased national sample. In addition, as the name implies, only surgical patients are included. A systematic sampling of cases from each member hospital is included in the data set each year. The most recent data file (from 2014) contains information on 750,397 cases from 517 participating sites. Annual data is available from 2005-2014 and includes sociodemographic information, procedure codes, comorbidities, preoperative labs, and certain operative details.
NIS and NSQIP in the Plastic Surgery Literature:
Systematic review of NIS/NSQIP studies in plastic surgery literature demonstrated 91 relevant citations (40 NIS, 51 NSQIP), with several trends readily apparent. First, there has been a dramatic recent increase in database studies [Fig 1]. 78% (31/40) of NIS and 100% (51/51) of NSQIP studies have been published within the past 4 years. Second, certain topics are more popular than others [Fig 2], with over half involving breast surgery. Third, single institutions frequently publish multiple studies on related topics [Fig 3]. Memorial Sloan Kettering (7) and UC Irvine (7) are responsible for over a third of all NIS publications, while Penn (21) and Northwestern (17) have combined for a staggering 75% of plastic surgery NSQIP papers.
Figure 1: Number of plastic surgery publications from NIS and NSQIP data by year.
Figure 2: Breakdown of NIS/NSQIP plastic surgery publications by topic of interest.
Figure 3: Breakdown of NIS/NSQIP publications by institution. The “Other” group consisted of 17 institutions with two or fewer publications.
Common Study Designs:
Two common study paradigms are used for NIS/NSQIP datasets. The first investigates the influence of patient and/or hospital characteristics on given outcomes. Common outcomes utilized include procedure type, complications, and cost. Predictive variables include comorbidities, demographics, and hospital characteristics. The most important analysis to scrutinize is the multivariate regression, where correlation of each variable with the outcome of interest is assessed while controlling for other variables.
A study utilizing this paradigm evaluated the likelihood of receiving autologous vs. implant-based breast reconstruction.4 The investigators found that age (50-59 years vs. younger), African-American race, delayed reconstruction, low-income, teaching hospitals, small hospitals, northeast region, and private insurance were associated with a greater likelihood of autologous reconstruction. They also found that delayed reconstruction, teaching hospitals, and private insurance were associated with greater likelihood of free flap than pedicle flap within the autologous subset.
The second study paradigm looks at trends in a given outcome over time. Common outcomes utilized are again procedures and complications. Although less commonly utilized, this is likely the more appropriate use of these datasets. Pien et al.5 evaluated rates of various flaps in autologous breast reconstruction from 2009-2011. They found no significant change in the total number of flap procedures over time, but DIEP flaps increased and LD flaps decreased in number, with no difference in pedicled or free TRAM flaps.
Strengths & Weaknesses:
The NIS/NSQIP databases have several strengths as research tools. They are publicly available on a year-by-year basis without IRB approval enabling rapid transition from generating a research question to performing statistical analyses – rarely possible in a single institution. Large sample size provides the power to detect small differences in outcomes and enables investigation of rare diseases. Finally, data included in the NIS is nationally representative, including subjects from all different hospital and payer types. This cohort heterogeneity may be difficult to come by in single center investigations.
It is equally important to consider the weaknesses of studies generated from these databases. First, both data sets are limited in their follow-up of included patients. NIS only contains data regarding isolated admissions with no information after discharge. The NSQIP, on the other hand, contains from the hospital stay and for thirty days following discharge. Second, given the retrospective nature of the data, it is not possible to prove causality. Ultimately the best possible conclusion is that a strong correlation exists between two variables while controlling for confounders as thoroughly as possible. Third, database size – while usually advantageous – can also become problematic. Statistical significance may not have clinical importance; always evaluate the absolute difference in outcomes to properly interpret results. Finally, there will always be some degree of coding error, which is particularly important to consider in the year or two following major coding changes within the ICD system.
The NIS and NSQIP databases have grown more popular every year as plastic surgery research tools. Indeed, these data sets can be tremendously useful for investigating a wide variety of research outcomes. It has also become apparent that one good research question may lead to multiple publications on related findings, demonstrating the importance of keeping an open mind when evaluating these databases. At the same time, however, it is paramount to recognize the limitations of these data sets when designing studies and interpreting results.
1. 2013 Introduction to the NIS. Healthcare Cost and Utilization Project (HCUP). November 2015. Agency for Healthcare Research and Quality, Rockville, MD. www.hcup-us.ahrq.gov/db/nation/nis/NIS_Introduction_2013.jsp
2. Khuri, Shukri F. “The NSQIP: a new frontier in surgery.” Surgery 138.5 (2005): 837-843.
3. Ko, C. “ACS NSQIP Conference and Semiannual Report Overview.” Presentation as the 2009 ACS NSQIP National Conference. 2009.
4. Albornoz CR, Bach PB, Pusic AL, McCarthy CM, Mehrara BJ, Disa JJ, Cordeiro PG, Matros E. The influence of sociodemographic factors and hospital characteristics on the method of breast reconstruction, including microsurgery: A US population–based study. Plast Reconstr Surg. 2012;129(5): 1071-1079.
5. Pien I, Caccavale S, Cheung MC, Butala P, Hughes DB, Ligh C, Zenn MR, Hollenbeck ST. Evolving trends in autologous breast reconstruction. Ann Plast Surg. 2014 Aug 28. [Epub ahead of print].