Accepted Manuscript

Big Data and Total Hip Arthroplasty: How Do Large Databases Compare?

Nicholas A. Bedard, M.D., Andrew J. Pugely, M.D., Michael A. McHugh, B.S., Nathan
R. Lux, Kevin J. Bozic, M.D., MBA, John J. Callaghan, M.D.

PII: S0883-5403(17)30786-6
DOI: 10.1016/j.arth.2017.09.003
Reference: YARTH 56077

To appear in: The Journal of Arthroplasty

Received Date: 20 July 2017

Revised Date: 28 August 2017
Accepted Date: 5 September 2017

Please cite this article as: Bedard NA, Pugely AJ, McHugh MA, Lux NR, Bozic KJ, Callaghan JJ, Big
Data and Total Hip Arthroplasty: How Do Large Databases Compare?, The Journal of Arthroplasty
(2017), doi: 10.1016/j.arth.2017.09.003.

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to
our customers we are providing this early version of the manuscript. The manuscript will undergo
copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please
note that during the production process errors may be discovered which could affect the content, and all
legal disclaimers that apply to the journal pertain.
 ACCEPTED MANUSCRIPT

Big Data and Total Hip Arthroplasty: How Do Large Databases Compare?

Nicholas A. Bedard, M.D.1

Andrew J. Pugely, M.D.1
Michael A. McHugh B.S.1
Nathan R. Lux1

PT
Kevin J. Bozic M.D., MBA2
John J. Callaghan, M.D.1

RI
Author Affiliation:

SC
1. The Department of Orthopaedic Surgery and Rehabilitation, University of Iowa Hospitals
and Clinics; Iowa City, Iowa
2. The Department of Surgery & Perioperative Care, Dell Medical School at the University

U
of Texas at Austin; Austin, Texas
AN
M

Corresponding Author:
D

Nicholas Bedard, MD
University of Iowa Hospitals and Clinics
TE

Department of Orthopaedics
200 Hawkins Drive, 01029 JPP
Iowa City, IA 52242
nicholas-bedard@uiowa.edu
C EP
AC
 ACCEPTED MANUSCRIPT

1 Big Data and Total Hip Arthroplasty: How Do Large Databases Compare?
2

PT
RI
U SC
AN
M
D
TE
C EP
AC

1
 ACCEPTED MANUSCRIPT

3 Abstract
4 Background: Use of large databases for orthopaedic research has become extremely popular in
5 recent years. Each database varies in the methods used to capture data and the population it
6 represents. The purpose of this study was to evaluate how these databases differed in reported
7 demographics, comorbidities and post-operative complications for primary total hip arthroplasty
8 patients.

PT
9
10 Methods: Primary THA patients were identified within National Surgical Quality Improvement
11 Programs (NSQIP), Nationwide Inpatient Sample (NIS), Medicare Standard Analytic Files

RI
12 (MED) and Humana administrative claims database (HAC). NSQIP definitions for comorbidities
13 and complications were matched to corresponding ICD-9/CPT codes to query the other
14 databases. Demographics, comorbidities, postoperative complications were compared.

SC
15
16 Results: The number of patients from each database was 22,644 in HAC, 371,715 in MED,
17 188,779 in NIS and 27,818 in NSQIP. Age and gender distribution were clinically similar.

U
18 Overall there was variation in prevalence of comorbidities and rates of postoperative
19 complications between databases. As an example, NSQIP had more than twice the obesity than
AN
20 NIS. HAC and MED had more than two times the diabetics than NSQIP. Rates of deep infection
21 and stroke 30-days after THA had more than twofold difference between all databases
22
M

23 Conclusions: Amongst databases commonly used in orthopaedic research, there is considerable

24 variation in complication rates following THA depending upon the database used for analysis. It
25 is important to consider these differences when critically evaluating database research.
D

26 Additionally, with the advent of bundled payments, these differences must be considered in risk
27 adjustment models.
TE

28
29 Keywords: Total hip arthroplasty, big data, database, clinical registry
30
C EP
AC

2
 ACCEPTED MANUSCRIPT

31 Introduction:
32
33 Over the last ten years, the use of big databases has increased substantially within the

34 field of orthopaedics (Figure 1) [1-3]. Clinical registries and administrative claims databases

35 have given researchers access to large cohorts of patients that allow for a more unique analysis of

PT
36 surgical complications, risk factors for such complications, trends in procedures and issues

RI
37 related to cost of care. Additionally, these large datasets are now being used for the public

38 reporting of surgical outcomes, with subsequent penalties for underperforming institutions [4, 5].

SC
39 Although similar in their ability to analyze large cohorts of patients and answer questions

40 difficulty to address with single institution type studies, each of these databases represent a

41
U
unique patient population, and vary in their methods of acquiring data [1-3]. Recent literature
AN
42 across multiple surgical specialties has begun to suggest that intrinsic differences in the methods

of data collection and the population of patients analyzed result in variability of reported
M

43

44 comorbidities and surgical complications [6-12]. Understanding the differences between these
D

45 databases is important for appropriately evaluating research utilizing them. Despite a significant
TE

46 amount of total hip arthroplasty (THA) literature being published utilizing these databases

47 (Figure 1), little has been done to evaluate the impact of a given databases on reported rates of
EP

48 comorbidities and complications following THA. Therefore, the purpose of this study was to

49 determine differences in reported demographics, comorbidities and complications for primary

C

50 THA patients amongst four commonly used databases.

AC

51

52 Methods:

53 Four large multicenter databases were utilized to complete this study: National Surgical

54 Quality Improvement Program (NSQIP), Nationwide Inpatient Sample (NIS), Medicare Standard

3
 ACCEPTED MANUSCRIPT

55 Analytic Files (MED) and Humana administrative claims database (HAC). Patients who had a

56 primary THA were identified using the International Classification of Diseases, 9th Revision

57 (ICD-9) code 81.51 and/or the Current Procedural Terminology (CPT) code 27130. Only

58 procedures performed between the years 2010 to 2012 were analyzed as these were the years

PT
59 common to all datasets. All four databases are Health Insurance Portability and Accountability

RI
60 Act (HIPAA) compliant and were found to be exempt from institutional review board approval

61 by our institutions Human Subjects Office.

SC
62 Three of the four databases utilized (HAC, MED, NIS) are administrative claims

63 databases and thus the data included in them are defined by reimbursement data linked to ICD-9

64
U
and/or CPT codes. The HAC database represents approximately 20 million lives and includes
AN
65 both privately insured patients and those who have purchased their Medicare/Medicaid

Advantage plans through Humana Inc. The MED database represented approximately 51 million
M

66

67 covered lives and is derived from Medicare parts A and B. Both of these databases capture
D

68 inpatient and outpatient events and utilized both CPT and ICD-9 codes. The NIS consists of a
TE

69 20% sample of all inpatient discharges and includes only inpatient data. Similar to HAC and

70 MED, NIS comorbidity and adverse event data are defined by reimbursement data for the
EP

71 inpatient admission of interest; however, only ICD-9 codes are supported by NIS. The NSQIP

72 database is a clinical registry maintained by American College of Surgeons that collects

C

73 comorbidities, procedure details and post-operative complications for patients undergoing major
AC

74 surgical procedures [13]. This database captures both inpatient and outpatient events occurring

75 within thirty-days following the surgical procedure of interest and uses trained clinical reviewers

76 to perform data collection through chart review and patient/surgeon contact utilizing strict

77 definitions for each comorbidity and complication variable cataloged (Table 1) [13].

4
 ACCEPTED MANUSCRIPT

78 Definitions from the NSQIP user manual for seven comorbidities (morbid obesity,

79 obesity, coagulopathy, diabetes, hypertension, chronic obstructive pulmonary disease (COPD),

80 and smoking) and nine postoperative complications (deep vein thrombosis (DVT), pneumonia,

81 stroke, myocardial infarction, cardiac arrest, pulmonary embolism (PE), wound dehiscence, deep

PT
82 surgical site infection (SSI), and any SSI) were matched to corresponding ICD-9 and CPT codes

RI
83 (Appendix 1 and 2). The PearlDiver Research Program (www.pearldiverin.com; PearlDiver Inc,

84 Fort Wayne, IN) was then used to query the three administrative claims databases (HAC, MED,

SC
85 NIS) using these ICD-9 and CPT codes. Demographic variables of age at the time of THA and

86 sex were obtained from each database.

87
U
The prevalence of complications following THA that occurred during the hospital stay
AN
88 were compared across all four databases. Given that wound dehisce following primary THA

occurred so infrequently during the inpatient stay we were not able compare the comorbidity
M

89

90 variable as HIPAA compliant databases are unable to report exact numbers when the value of the
D

91 endpoint of interest lies between zero and ten. Complications that occurred within thirty-days
TE

92 following the primary THA procedure were compared amongst databases with that interval of

93 follow-up available (HAC, MED, NSQIP). Lastly, the differences in rates of complications
EP

94 occurring during the inpatient stay versus those occurring within thirty-days following surgery

95 were compared for databases with both of these times points available (HAC, MED, NSQIP).
C

96 Pearson chi-square test was utilized to compare demographic variables with a p value of
AC

97 <0.05 considered significant. Relative risk (RR) and corresponding 95% confidence intervals

98 were utilized to compare the prevalence of comorbidities and complications between databases.

99 However, consistent with other database comparison studies, the large sample sizes and

100 associated high power allowed for detection of statistical significance for small, clinically

5
 ACCEPTED MANUSCRIPT

101 insignificant differences [6, 7]. Thus, the focus of these comparisons was on the magnitude of the

102 differences, specifically utilizing an absolute difference threshold of greater than twofold to

103 signify an important clinical difference.

104

PT
105 Results:

RI
106 The number of primary THA patients analyzed in each database was: 22,644 in HAC,

107 371,715 in MED, 188,779 in NIS and 27,818 in NSQIP. The female to male ratio was 1.3:1 for

SC
108 NIS and NSQIP, 1.6:1 for MED and 1.4:1 for HAC (p <0.001). The age group distribution

109 between databases is seen in Figure 2 and comparisons at each age groups were statistically

110
U
significant (p<0.001). The median age at time of THA fell within the 70-74 year old age group
AN
111 HAC, 75-79 year old group for MED and 65-69 year old age group for NIS and NSQIP (Figure

2).
M

112

113 Analysis of prevalence of comorbidities and post-operative complications demonstrated

D

114 much variability between datasets (Figure 3, 4 and 5). Relative risk values and corresponding
TE

115 95% confidence intervals comparing these endpoints amongst each database can be found in in

116 Tables 2, 3 and 4. However, given that most comparison were statistically significant in the
EP

117 relative risk analysis despite many small clinically insignificant differences, the focus of these

118 comparisons is on the absolute difference in comorbidities and complication rates using a
C

119 threshold of twofold difference to determine clinical significance.

AC

120 Comparison of comorbidities amongst patients in each database demonstrated HAC and

121 MED to have twice the prevalence of COPD, coagulopathy and diabetes than NSQIP. Similarly,

122 HAC and MED had over two times the prevalence of COPD and coagulopathy as NIS. Lastly,

6
 ACCEPTED MANUSCRIPT

123 NSQIP had over twice the prevalence of obesity as NIS. All other comorbidity comparison were

124 not different between databases (less than two-fold difference).

125 Prevalence of complications occurring during the inpatient stay for primary THA

126 revealed HAC to have over twice the amount of complications than NSQIP for all variables

PT
127 analyzed. HAC had greater than two-fold prevalence of DVT and any SSI than MED and NIS.

RI
128 Additionally, HAC had over two times the amount of inpatient strokes as NIS. MED had more

129 than twice the amount of DVT, pneumonia, stroke, MI, cardiac arrest and any SSI as NSQIP.

SC
130 Lastly, NIS had over two-fold prevalence of pneumonia during the inpatient stay as NSQIP. All

131 other comparisons of inpatient complications were within two-fold difference between databases

132 (Figure 4).

U
AN
133 Prevalence of complications occurring within thirty-days following THA demonstrated

HAC to have greater than two-fold prevalence of DVT, stroke and deep SSI compared to MED.
M

134

135 Additionally, HAC had more than twice the prevalence of all complication endpoints analyzed
D

136 relative to NSQIP. Comparing MED to NSQIP, revealed MED to have over two times the
TE

137 amount of pneumonia, stroke, MI, cardiac arrest, PE and wound dehiscence as NSQIP.

138 Conversely, NSQIP had more than twice the prevalence of deep SSI as MED. Prevalence of all
EP

139 other complications occurring within thirty-days of surgery were not different between databases

140 (Figure 5).

C

141 When comparing the number of complications captured by a given database relative to
AC

142 the duration of follow-up included (inpatient versus thirty-days following surgery), it was found

143 that over half of any SSI, deep SSI and DVTs are not captured by any database (NSQIP, HAC,

144 MED) if follow-up is limited to inpatient events only. Additionally, all complication variables at

145 least doubled in prevalence in the HAC database when follow-up was extended from inpatient

7
 ACCEPTED MANUSCRIPT

146 only to thirty-days following THA. Lastly, over half of PEs were not captured by MED if only

147 inpatient data was analyzed.

148

149 Discussion:

PT
150 As use of large databases continues to increase for orthopaedic research and assessments

RI
151 of hospital quality, this study demonstrates that amongst four commonly utilized databases there

152 is considerable variation in prevalence of comorbidities and complications following primary

SC
153 THA. These findings highlight the importance of understanding the methodology utilized build

154 each database and the population that is being captured for that dataset.

155
U
We suspect that NSQIP, which utilized chart abstraction with strict definitions for each
AN
156 variable, is likely more accurate than administrative claims databases given limitations in ICD-

9/CPT codes and the reliance on accurate documentation. The suggestion that administrative
M

157

158 claims data may be less accurate than clinical registries has recently begun to emerge within the
D

159 literature across multiple surgical specialties [6-12]. Two studies within orthopaedics performed
TE

160 a similar analysis as our present study for patients who underwent lumbar spine surgery and for

161 hip fracture patients [6, 7]. They similarly found variability in prevalence of complications
EP

162 between NSQIP and NIS for each of the surgical procedures analyzed. Additionally, they also

163 cited the impact analysis of only the impatient stay (compared to thirty-day follow-up) has on the
C

164 reported prevalence of complications after hip fracture and lumbar spine surgery [6, 7]. One
AC

165 strength the present study has over these analyses, is that multiple administrative claims

166 databases (MED, HAC, NIS) were compared amongst themselves and to NSQIP. This allowed

167 us to concluded that despite the data cited above, the differences in prevalence of complications

168 between databases cannot be entirely attributed to the type of database (administrative claims

8
 ACCEPTED MANUSCRIPT

169 versus clinical registry) as even comparisons between multiple administrative claims databases

170 in this study had drastically different rates of surgical complications (Figure 4 and 5) despite

171 querying the database with the exact same methodology and coding algorithms.

172 Despite these results, we are not suggesting that databases that only include the inpatient

PT
173 stay or administrative claims databases have no role in the orthopaedic literate. Administrative

RI
174 claims databases such as HAC and MED are often more useful for evaluation of longer term

175 complications as they are not limited to a specific time period of analysis (ie 30 days

SC
176 postoperatively) and patients can be tracked the time period of the databases for specific events.

177 Evaluation of trends overtime and analysis of prevalence of a given intervention or complication

178
U
seem to be best asses with databases such NIS, MED and HAC as they are more representative
AN
179 of the general population, patients are able to be tracked overtime and they generally have more

years of data included. Further administrative claims databases are better positioned for financial
M

180

181 analysis as they are based upon reimbursement data. Additionally, although inpatient databases
D

182 do not capture many short term complications that occur soon after discharge, they can be very
TE

183 important for evaluation of issues related to length of stay or inpatient events. Lastly, analysis of

184 questions outside the scope of strict clinical definitions often associated with clinical registries is
EP

185 better studied with administrative claims databases using diagnosis and procedure codes to

186 define outcomes not included in a given clinical registry. Clinical registries like NQSIP seem to
C

187 be best suited for evaluation of specific comorbidities on given complications after surgical
AC

188 procedures given their strictly defined definitions for variables included and the prospective

189 collection of data by trained clinical reviewers. We believe these results are useful to highlight

190 the discrepancies that can occur amongst multiple different databases being used to answer the

191 same question in order to emphasize the importance of understanding the strengths and

9
 ACCEPTED MANUSCRIPT

192 weaknesses to the methodology used for data acquisition when evaluating literature that utilizes

193 a large administrative claims database or a clinical registry.

194 Limitations to this study include the inability to identify a specific patient across multiple

195 databases, which would have allowed for true measure of validity and determination of the most

PT
196 accurate database for analyzing THA complications. Additionally, due to the limitations of ICD-

RI
197 9/CPT coding and strict NSQIP definitions of comorbidities and complications, it is likely that

198 the results may be biased by imperfect matching for some variables. Lastly, the two-fold

SC
199 threshold used to signify an excessive clinically important difference may have missed smaller,

200 but still clinically significant, differences in the prevalence of comorbidities and complications

201
U
following THA. This would have been more likely to occur for comorbidities and complications
AN
202 with a higher baseline prevalence; as this would have needed to double in order to meet the two-

fold cutoff. However, it was felt that having a higher cutoff threshold was imperative to highlight
M

203

204 the large clinically important differences within these datasets.

D

205 In conclusion, this study demonstrates that amongst clinical and administrative databases
TE

206 commonly used in orthopaedic research, there is considerable variation in prevalence of

207 comorbidities and rates of complication following primary THA depending upon the database
EP

208 and postoperative time period used for analysis. It will be important to consider these differences

209 when critically evaluating research utilizing large databases. Additionally, in the era of value
C

210 based healthcare, these differences must be considered when using large databases to develop
AC

211 risk adjustment models for comparing provider performance.

212

10
 ACCEPTED MANUSCRIPT

213 References:
214
215 1. Pugely AJ, Martin CT, Harwood J, Ong KL, Bozic KJ, Callaghan JJ. Database and
216 Registry Research in Orthopaedic Surgery: Part I: Claims-Based Data. J Bone Joint Surg
217 Am 2015;97(15):1278-87.
218 2. Pugely AJ, Martin CT, Harwood J, Ong KL, Bozic KJ, Callaghan JJ. Database and

PT
219 Registry Research in Orthopaedic Surgery: Part 2: Clinical Registry Data. J Bone Joint
220 Surg Am 2015;97(21):1799-808.
221 3. Bohl DD, Singh K, Grauer JN. Nationwide Databases in Orthopaedic Surgery Research. J
222 Am Acad Orthop Surg 2016;24(10):673-82.

RI
223 4. US Senate. HR 3590: The Patient Protection and Affordable Care Act. 2009
224 https://www.govtrack.us/congress/bills/111/hr3590/text. Accessed 2016 August 25
225 5. Centers for Medicare & Medicaid Services. CMS dry run hospital-specific report for

SC
226 hospital-wide all-cause unplanned readmission (HWR) measure. 2012
227 http://www.qualitynet.org. Accessed 2016 August 15
228 6. Bohl DD, Basques BA, Golinvaux NS, Baumgaertner MR, Grauer JN. Nationwide

U
229 Inpatient Sample and National Surgical Quality Improvement Program give different
230 results in hip fracture studies. Clin Orthop Relat Res 2014;472(6):1672-80.
AN
231 7. Bohl DD, Russo GS, Basques BA, Golinvaux NS, Fu MC, Long WD, 3rd, Grauer JN.
232 Variations in data collection methods between national databases affect study results: a
233 comparison of the nationwide inpatient sample and national surgical quality improvement
234 program databases for lumbar spine fusion procedures. J Bone Joint Surg Am
M

235 2014;96(23):e193.
236 8. Awad MI, Shuman AG, Montero PH, Palmer FL, Shah JP, Patel SG. Accuracy of
237 administrative and clinical registry data in reporting postoperative complications after
D

238 surgery for oral cavity squamous cell carcinoma. Head Neck 2015;37(6):851-61.
239 9. Enomoto LM, Hollenbeak CS, Bhayani NH, Dillon PW, Gusani NJ. Measuring surgical
TE

240 quality: a national clinical registry versus administrative claims data. J Gastrointest Surg
241 2014;18(8):1416-22.
242 10. Kulaylat AN, Engbrecht BW, Rocourt DV, Rinaldi JM, Santos MC, Cilley RE,
243 Hollenbeak CS, Dillon PW. Measuring Surgical Site Infections in Children: Comparing
EP

244 Clinical, Electronic, and Administrative Data. J Am Coll Surg 2016;222(5):823-30.

245 11. Lawson EH, Louie R, Zingmond DS, Brook RH, Hall BL, Han L, Rapp M, Ko CY. A
246 comparison of clinical registry versus administrative claims data for reporting of 30-day
C

247 surgical complications. Ann Surg 2012;256(6):973-81.

248 12. Lawson EH, Louie R, Zingmond DS, Sacks GD, Brook RH, Hall BL, Ko CY. Using
AC

249 Both Clinical Registry and Administrative Claims Data to Measure Risk-adjusted
250 Surgical Outcomes. Ann Surg 2016;263(1):50-7.
251 13. American College of Surgeons National Surgical Quality Improvement Program. Data
252 Collection, Analysis, and Reporting. . http://site.acsnsqip.org/program-specifics/data-
253 collection-analysis-and-reporting/. Accessed 2016 May 15
254
255

11
 ACCEPTED MANUSCRIPT

256 Figure legend:

257
258 Figure 1: Number of database publications found in PubMed with the following search query:
259 (("total hip replacement" or "total hip arthroplasty" or "knee replacement" or "THA" or "total
260 hip" or "hip replacement")) AND ("NIS" or "nationwide inpatient sample" or "NSQIP" or
261 "national surgical quality improvement program" or "PearlDiver" or "national database" or
262 "insurance database" or "database")

PT
263
264 Figure 2: Breakdown of age of the patients at time of primary THA
265

RI
266 Figure 3: Prevalence of comorbidities in primary THA patients
267
268 Figure 4: Prevalence of inpatient complications following THA for each database

SC
269
270 Figure 5: Prevalence of complications within thirty-days of THA for each database
271

U
AN
M
D
TE
C EP
AC

12
 ACCEPTED MANUSCRIPT

Table 1: Comparison of Databases

HAC MED NIS NSQIP
Patient data Retrospective review of Retrospective review of Retrospective review of Prospective samples of
patients insured through patients insured through a national samples of patients from a clinical

PT
Humana Medicare parts A and B inpatient discharges registry
Definition of Procedures ICD-9/CPT codes ICD-9/CPT codes ICD-9 codes CPT codes
Definition of ICD-9 codes associated ICD-9 codes associated ICD-9 codes associated Medical records reviews

RI
Comorbidities/complications with reimbursement data with reimbursement data with reimbursement data by trained reviewers using
strict clinical definitions

SC
Time period included Inpatient and outpatient Inpatient and outpatient Inpatient only Inpatient and outpatient
for 30 days
postoperatively

U
HAC: Humana administrative claims database, MED: Medicare Standard Analytic Files,NIS: National Inpatient Sample, NSQIP:

AN
National Surgical Quality Improvement Program , ICD-9: International Classification of Diseases, 9th Revision, CPT: Current
Procedural Terminology.

M
D
TE
C EP
AC
 ACCEPTED MANUSCRIPT

Table 2: Comparisons of Comorbidity Prevalence Amongst THA Patients

Comorbidity HAC vs MED HAC vs NIS HAC vs NSQIP MED vs NIS MED vs NSQIP NIS vs NSQIP
Smoking 1.12 (1.08-1.16) 1.03 (0.99-1.07) 0.81 (0.77-0.85) 0.92 (0.91-0.94) 0.72 (0.70-0.75) 0.78 (0.76-0.81)
COPD 1.21 (1.18-1.24) 3.48 (3.38-3.58) 5.32 (5.00-5.65) 2.87 (2.82-2.92) 4.39 (4.14-4.64) 1.53 (1.44-1.62)

PT
HTN 1.03 (1.02-1.03) 1.48 (1.46-1.49) 1.42 (1.40-1.43) 1.44 (1.43-1.45) 1.38 (1.37-1.40) 0.96 (0.95-0.97)
Diabetes 1.12 (1.1-1.15) 1.94 (1.90-1.99) 2.54 (1.45-2.64) 1.73 (1.71-1.76) 2.27 (2.19-2.34) 1.31 (1.27-1.36)

RI
Coagulopathy 1.00 (0.96-1.05) 3.12 (2.96-3.29) 2.82 (2.59-3.06) 3.11 (3.02-3.21) 2.96 (2.75-3.18) 0.90 (0.83-0.97)
Obesity 1.16 (1.13-1.19) 1.83 (1.78-1.88) 0.60 (0.59-0.62) 1.58 (1.56-1.60) 0.52 (0.51-0.53) 0.33 (0.32-0.34)
Morbid Obesity 1.24 (1.19-1.29) 1.54 (1.48-1.62) 1.30 (1.22-1.38) 1.25 (1.23-1.28) 1.05 (1.00-1.10) 0.87 (0.83-0.91)

SC
Data presented as relative risk and corresponding 95% confidence interval. HAC: Humana Administrative Claims database, MED:
Medicare Standard Analytic Files, NIS: National Inpatient Sample, NSQIP: National Surgical Quality Improvement Program, COPD:

U
chronic obstructive pulmonary disease, HTN: hypertension.

AN
M
D
TE
C EP
AC
 ACCEPTED MANUSCRIPT

Table 3: Comparisons of Inpatient Complications following THA

Complication HAC vs MED HAC vs NIS HAC vs NSQIP MED vs NIS MED vs NSQIP NIS vs NSQIP
DVT 3.69 (2.53-5.37) 2.52 (2.03-3.12) 3.69 (2.53-5.37) 1.46 (1.29-1.64) 2.14 (1.53-2.98) 1.47 (1.04-2.06)
Pneumonia 1.17 (1.02-1.34) 1.79 (1.55-2.07) 4.61 (3.47-6.12) 1.53 (1.43-1.64) 3.94 (3.05-5.09) 2.57 (1.98-3.33)

PT
Stroke 1.88 (1.51-2.33) 2.67 (2.11-3.78) 4.52 (2.91-7.03) 1.42 (1.25-1.63) 2.41 (1.62-3.59) 1.69 (1.12-2.55)
MI 1.04 (0.87-1.26) 1.51 (1.24-1.85) 2.41 (1.76-3.31) 1.45 (1.32-1.59) 2.31 (1.78-3.00) 1.60 (1.22-2.09)

RI
Cardiac Arrest 1.13 (0.78-1.62) 1.87 (1.27-2.76) 3.40 (2.57-6.09) 1.66 (1.37-2.01) 1.69 (1.08-2.64) 1.02 (0.64-1.62)
PE 1.37 (1.09-1.73) 1.96 (1.52-2.52) 2.39 (1.62-3.52) 1.43 (1.26-1.62) 1.75 (1.27-2.41) 1.22 (0.87-1.70)
Any SSI 3.30 (2.78-3.91) 4.00 (3.32-4.84) 5.14 (3.61-7.32) 1.22 (1.07-1.38) 5.14 (3.61-7.32) 1.28 (0.91-1.80)

SC
Data presented as relative risk and corresponding 95% confidence interval. HAC: Humana Administrative Claims database, MED:
Medicare Standard Analytic Files, NIS: National Inpatient Sample, NSQIP: National Surgical Quality Improvement Program, DVT:

U
deep vein thrombosis, MI: myocardial infarction, PE: pulmonary embolism, SSI: surgical site infection.

AN
M
D
TE
C EP
AC
 ACCEPTED MANUSCRIPT

Table 4: Comparisons of Complications within Thirty-Days Following THA

Complication HAC vs MED HAC vs NSQIP MED vs NSQIP

PT
DVT 2.34 (2.17-2.53) 7.25 (6.01-8.74) 1.98 (1.66-2.36)
Pneumonia 1.88 (1.73-2.03) 8.47 (6.84-10.5) 4.52 (3.69-5.52)
Stroke 2.93 (2.57-3.34) 7.58 (5.47-10.5) 2.59 (1.90-3.52)

RI
MI 1.42 (1.24-1.63) 3.37 (2.63-4.34) 2.37 (1.91-2.95)
Cardiac Arrest 1.86 (1.49-2.32) 3.96 (2.57-6.09) 2.13 (1.91-2.95)

SC
PE 1.86 (1.63-2.11) 4.70 (3.60-6.13) 2.53 (1.98-3.21)
Wound Dehiscence 1.55 (1.28-1.86) 4.64 (3.15-6.85) 3.00 (2.11-4.26)
Deep SSI 7.08 (6.09-8.23) 2.11 (1.72-2.60) 0.30 (0.25-0.36)

U
Any SSI 1.98 (1.83-2.13) 2.57 (2.28-2.91) 1.30 (1.17-1.45)

AN
Data presented as relative risk and corresponding 95% confidence interval. HAC: Humana Administrative Claims database, MED:
Medicare Standard Analytic Files, NSQIP: National Surgical Quality Improvement Program, DVT: deep vein thrombosis, MI:
myocardial infarction, PE: pulmonary embolism, SSI: surgical site infection.

M
D
TE
C EP
AC
 ACCEPTED MANUSCRIPT

PT
RI
U SC
AN
M
D
TE
EP
C
AC
 ACCEPTED MANUSCRIPT

PT
RI
U SC
AN
M
D
TE
EP
C
AC
 ACCEPTED MANUSCRIPT

PT
RI
U SC
AN
M
D
TE
EP
C
AC
 ACCEPTED MANUSCRIPT

PT
RI
U SC
AN
M
D
TE
EP
C
AC
 ACCEPTED MANUSCRIPT

PT
RI
U SC
AN
M
D
TE
EP
C
AC
 ACCEPTED MANUSCRIPT

Appendix 1. Definitions of Comorbidity Variables

Comorbidity NSQIP NIS, HAC, MED
Morbid Obesity BMI > 40 kg/m2 based upon “height” ICD-9 278.01 (Obesity [BMI from >40 kg/m2 ) V85.3 (BMI over

PT
(height) and “weight” (weight) 40 kg/m2, Adult)
Obesity BMI 30-40 kg/m2 based upon “height” ICD-9 278.00 (Obesity [BMI from 30-40 kg/m2 ) V85.3 (BMI
(height) and “weight” (weight) between 30-39 kg/m2, Adult)

RI
Coagulopathy “bleeddis” (bleeding disorders) ICD-9 286.x (coagulation defect), 287.3x (primary
thrombocytopenia), 287.4x (secondary thrombocytopenia), 287.5

SC
Thrombocytopenia, unspecified), 269.0 (deficiency of vitamin K)
Diabetes “diabetes” (diabetes mellitus with oral ICD-9 249.xx (secondary diabetes mellitus), 250.xx (diabetes
agents or insulin) mellitus)

U
Hypertension “hypermed” (hypertension requiring ICD-9 401.xx (essential hypertension), 405.xx (secondary

AN
medication) hypertension)
Chronic Obstructive “hxcopd” (history of severe COPD) ICD-9 491.xx (chronic bronchitis), 492.xx (emphysema),496
Pulmonary disease (chronic airway obstruction, unspecified)

M
Smoking “smoke” (current smoker within one ICD-9 30.51 (tobacco use disorder, dependence)
year)

D
NSQIP = National Surgical Quality Improvement Program, NIS = National Inpatient Sample, MED = Medical Standard Analytic
Files, HAC = Humana Administrative Claims database. ICD-9 = International Classification of Diseases, 9th Revision. The letter “x”

TE
represents all fourth or fifth digits that could designate an ICD-9 code.
EP
C
AC
 ACCEPTED MANUSCRIPT

Appendix 2. Definitions of Complication Variables

Comorbidity NSQIP HAC and MED NIS
Deep Vein “nothdvt” (deep vein Acute venous embolism and Acute venous embolism and thrombosis of:
Thrombosis thrombosis, with or thrombosis of: ICD453.2 ICD453.2 (inferior vena cava), 453.3 (renal

PT
without (inferior vena cava), 453.3 (renal vein), 453.40 (deep vessel unspecified),
inflammation, vein), 453.40 (deep vessel 453.41 (proximal lower extremity), 453.42
postoperative) unspecified), 453.41 (proximal (distal lower extremity), 453.82 (deep

RI
lower extremity), 453.42 (distal veins of upper extremity), 453.83 (upper
lower extremity), 453.82 (deep extremity unspecified), 453.84 (axillary

SC
veins of upper extremity), 453.83 vein), 453.85 (subclavian vein), 453.86
(upper extremity unspecified), (internal jugular vein)
453.84 (axillary vein), 453.85

U
(subclavian vein), 453.86

AN
(internal jugular vein)
Pneumonia “noupneumo” ICD-9 480.x (viral pneumonia), ICD-9 480.x (viral pneumonia), 481
(pneumonia, 481 (pneumococcal pneumonia), (pneumococcal pneumonia), 482.x (other

M
postoperative) 482.x (other bacterial bacterial pneumonia), 483.x (pneumonia
pneumonia), 483.x (pneumonia due to other specified organism), 484.x
due to other specified organism), (pneumonia in infectious diseases

D
484.x (pneumonia in infectious classified elsewhere), 485

TE
diseases classified elsewhere), (bronchopneumonia), 486 (pneumonia,
485 (bronchopneumonia), 486 organism unspecified)
(pneumonia, organism
EP
unspecified)
Stroke “ncnscva” (stroke or ICD-9 997.02 (iatrogenic ICD-9 997.02 (iatrogenic cerebrovascular
cerebrovascular accident, cerebrovascular infarction or infarction or hemorrhage), 430 (SAH), 431
C

postoperative) hemorrhage), 430 (SAH), 431 (intracerebral hemorrhage), 432.x (other

AC

(intracerebral hemorrhage), 432.x and unspecified intracranial hemorrhage),

(other and unspecified
intracranial hemorrhage), 433.x1
(occlusion and stenosis of
precerebral arteries with
infarction), 434.x1 (occlusion of
433.x1 (occlusion and stenosis of
precerebral arteries with infarction), 434.x1
(occlusion of cerebral arteries, with
cerebral infarction)
 ACCEPTED MANUSCRIPT

cerebral arteries, with cerebral

infarction)
Myocardial Infarction “ncdmi” (myocardial ICD-9 410.x0, 410.x1 (acute ICD-9 410.x0, 410.x1 (acute myocardial
infarction, postoperative) myocardial infarction, initial and infarction, initial and unspecified episode

PT
“typeintoc” (myocardial unspecified episode of care) of care)
infarction, intraoperative)
Cardiac Arrest “ncdarrest” (cardiac arrest ICD-9 427.5 (Cardiac arrest), ICD-9 427.5 (Cardiac arrest), 427.41

RI
requiring 427.41 (Ventricular fibrillation), (Ventricular fibrillation)
cardiopulmonary CPT-92950 (Cardiopulmonary

SC
resuscitation, resuscitation; eg, in cardiac
postoperative) arrest)
“typeintoc” (cardiac

U
arrest requiring

AN
cardiopulmonary
resuscitation,
intraoperative)

M
Pulmonary Embolism “npulembol” (pulmonary ICD-9 415.11 (atrogenic ICD-9 415.11 (atrogenic pulmonary
embolism, postoperative) pulmonary embolism and embolism and infarction), 415.12 (septic
infarction), 415.12 (septic pulmonary embolism), 415.13 (Saddle

D
pulmonary embolism), 415.13 embolus of pulmonary artery), 415.19

TE
(Saddle embolus of pulmonary (Other pulmonary embolism)
artery), 415.19 (Other pulmonary
embolism)
EP
Wound Dehiscence “ndehis” (wound ICD-9 998.30 (disruption of ICD-9 998.30 (disruption of wound,
disruption, postoperative) wound, unspecified), 998.31 unspecified), 998.31 (disruption of an
(disruption of an internal internal operation (surgical) wound),
C

operation (surgical) wound), 998.32 (disruption of an external operation

AC

998.32 (disruption of an external (surgical) wound)

operation (surgical) wound)
Deep Surgical Site “nwndinfd” (deep CPT 27091 (Removal of hip NA*
Infection surgical site infection, prosthesis; complicated,
postoperative); including total hip prosthesis,
norgspcssi” (organ or methylmethacrylate with or
 ACCEPTED MANUSCRIPT

space without insertion of spacer),

surgical site infection, 10180 (incision and drainage,
postoperative) complex, postoperative wound
infection), 27030 (Arthrotomy,

PT
hip, with drainage (ie infection),
26990 (incision and drainage,
pelvis or hip joint area, deep

RI
abcess or hematoma), 26991
(incision and drainage, pelvis or

SC
hip joint area, infected bursa),
11981 (insertion, non-
biodegradable drug delivery

U
implant), ICD-9 80.05

AN
(Arthrotomy for removal of
prosthesis without replacement,
hip)

M
Any Surgical Site “nsupinfec” (superficial ICD-9 996.66 (infection and ICD-9 996.66 (infection and inflammatory
Infection surgical site infection, inflammatory reaction due to reaction due to internal joint prosthesis),
Postoperative)“nwndinfd” internal joint prosthesis), 998.83 (non-healing surgical wound),

D
(deep surgical site 998.83 (non-healing surgical 998.51 (infected postoperative seroma),

TE
infection, postoperative); wound), 998.51 (infected 998.59 (other postoperative infection),
“norgspcssi” (organ or postoperative seroma), 998.6 (persistent postoperative fistula),
space surgical site 998.59 (other postoperative 80.05 (Arthrotomy for removal of
EP
infection, postoperative) infection), 998.6 (persistent prosthesis without replacement, hip)
postoperative fistula)
C

CPT 27091 (Removal of hip

AC

prosthesis; complicated,
including total hip prosthesis,
methylmethacrylate with or
without insertion of spacer),
10180 (incision and drainage,
complex, postoperative wound
 ACCEPTED MANUSCRIPT

infection), 27030 (Arthrotomy,

hip, with drainage (ie infection),
26990 (incision and drainage,
pelvis or hip joint area, deep

PT
abcess or hematoma), 26991
(incision and drainage, pelvis or
hip joint area, infected bursa),

RI
11981 (insertion, non-
biodegradable drug delivery

SC
implant), ICD-9 80.05
(Arthrotomy for removal of
prosthesis without replacement,

U
hip)

AN
M
D
TE
EP
C
AC