Journal of Thrombosis and Haemostasis, 8: 987–997
DOI: 10.1111/j.1538-7836.2010.03798.x
ORIGINAL ARTICLE
Recurrent venous thromboembolism after surgery-provoked versus unprovoked thromboembolism R . H . W H I T E , * S . M U R I N , * T . W U N * and B . D A N I E L S E N *Department of Medicine, UC Davis School of Medicine; and Health Information Solutions, Sacramento, CA, USA
To cite this article: White RH, Murin S, Wun T, Danielsen B. Recurrent venous thromboembolism after surgery-provoked versus unprovoked thromboembolism. J Thromb Haemost 2010; 8: 987–97.
Introduction Summary. Background: The incidence of recurrent venous thromboembolism (VTE) varies depending on the nature of the initial provoking risk factor(s). Objectives: To compare the incidence and time course of recurrent VTE after unprovoked VTE vs. VTE provoked by nine different types of surgery. Methods: Retrospective analysis of linked California hospital and emergency department discharge records. Between 1997 and 2007, all surgery-provoked VTE cases had a first-time VTE event diagnosed within 60 days after undergoing a major operation. The incidence of recurrent VTE was compared during specified follow-up periods by matching each surgeryprovoked case with three unprovoked cases based on age, race, gender, VTE event, calendar year and co-morbidity. Results: The 4-year Kaplan–Meier cumulative incidence of recurrent VTE was 14.7% (95%CI: 14.2–15.1) in the matched unprovoked VTE group vs. 7.6% (CI: 7.0–8.2) in 11 797 patients with surgery-provoked VTE (P < 0.001). The overall risk reduction was 48%, which ranged from 64% lower risk (P < 0.001) after coronary bypass surgery to 25% lower risk (P = 0.06) after disc surgery. The risk of recurrent VTE 1– 5 years after the index event was significantly lower in the surgery group (HR = 0.47, CI: 0.41–0.53). Within the surgeryprovoked group, the risk of recurrent VTE was similar in men and women (HR = 1.0, CI: 0.8–1.3). Conclusions: The risk of recurrent VTE after surgery-provoked VTE was approximately 50% lower than after unprovoked VTE, confirming the view that provoked VTE is associated with a lower risk of recurrent VTE. However, there was appreciable heterogeneity in the relative risk of recurrent VTE associated with different operations. Keywords: epidemiology, outcomes, provoked, recurrent, surgery, unprovoked, venous thromboembolism. Correspondence: Richard H. White, Suite 2400, PSSSB 4150 V Street, Sacramento, CA 95817, USA. Tel.: + 1 916 734 7005; fax: + 1 916 734 2732; E-mail:
[email protected] Received 25 January 2010, accepted 28 January 2010 2010 International Society on Thrombosis and Haemostasis
Venous thromboembolism (VTE), which encompasses both lower extremity deep vein thrombosis (DVT) and pulmonary embolism (PE), is increasingly viewed as a chronic disease given that over 30% of all patients have a recurrent VTE event within 10 years [1–4], 1–3% develop thromboembolic pulmonary hypertension [5] and 5–40% develop post-thrombotic syndrome [6]. Several risk factors have been identified that are associated with a higher incidence of recurrent VTE [7], with one of the strongest being absence of an overt provocation for the initial thrombotic event, such as recent trauma or major surgery. Indeed, most studies of recurrent VTE stratify patients into two groups, those with a provoked (or ÔsecondaryÕ) VTE event and those with an unprovoked (or ÔidiopathicÕ) VTE event [3,7]. Management recommendations are based upon this categorization and they differ substantially for the two groups [8]. The strength and consistency of the tenet that patients who develop acute VTE after a provocation have a lower longterm incidence of recurrent VTE appears to be quite strong. However, if one looks at the incidence of acute VTE after major surgery, one sees a very large variation in the observed incidence of acute VTE after different operations [9]. This raises the possibility that there might indeed be significant differences in the incidence of recurrent VTE depending on the nature of the inciting operation. Unfortunately, because studies that have analyzed the effect of surgery-provoked VTE on the incidence of recurrent VTE have included relatively few patients, researchers have had to combine together all of the patients who developed VTE after any operation in order to create a Ôsurgery-provokedÕ VTE cohort [1,2,4,10–14]. Because only 0.5–2% of patients develop symptomatic VTE after major surgery [9], and only a modest percentage of these cases develop recurrent VTE [15], in order to compare the incidence of recurrent VTE after different operations one needs a very large database of surgery cases. Such a database exists in California, which has mandated the reporting of hospital discharge information from all public hospitals since 1991, and emergency department records since 2005. These
988 R. H. White et al
records can be linked together to analyze health encounters over time in this large geographic area. These data are utilized not just for reimbursement purposes, but also for quality review, benchmarking measurements and the collection of general medical statistical data [16]. Use of medical discharge records is a very valuable research tool because it is relatively inexpensive, is more reliable than patient or physician reporting, is superior to data from third party payers, covers large populations and is available over multiple years [17]. The principal weakness of this data is potential inconsistency in coding medical diagnoses across hospitals and the absence of specific data elements, such as radiology reports or use of anticoagulants. Utilizing linked hospital discharge data from the entire State of California, we hypothesized that among patients who developed acute VTE within 60 days after one of nine different types of major surgery, the incidence of subsequent recurrent VTE would vary, but that for all operations, that incidence of recurrent VTE would be lower after surgery-provoked VTE than after unprovoked VTE. We also hypothesized the women would have a lower risk of recurrent VTE than men [7]. Methods This was a retrospective observational study that compared the incidence of recurrent VTE among patients diagnosed with either a first-time VTE event within 60 days after one of nine specified types of operations performed between 1 January 1 1997 and 2 October 2 2007, or a first time unprovoked VTE during this same time period. Databases
California requires that all non-federal hospitals (1991–2007) and emergency departments (ED) (2005–2007) provide information about patient health encounters in its non-federal facilities. It is possible to link all of these records together using an encrypted form of the social security number together with several other variables using deterministic linkage procedures [18,19]. Fewer than 5% of records lacked a social security number or could not be linked. Follow-up based on linked records has been shown to be as effective as reporting based on direct contact with the patients [19,20]. The Patient hospital Discharge Database (PDD) and ED database include demographic information, the principal medical diagnosis (condition occasioning admission to the hospital), up to 24 secondary medical diagnoses, a principal procedure and up to 20 secondary procedures using International Classification of Diseases, 9th Revision, Clinical Modification (ICD-9-CM) codes. Since 1996 all medical diagnoses listed in the PDD have been indexed as being either present at the time of admission (POA = Y) or not (POA = N), with some diagnoses indexed as unclear (POA = M) or unknown (POA = U). This study was approved by the California Health and Welfare Agency Committee for the Protection of Human Subjects, and the University of California, Davis Institutional Review Board.
Definition of index and recurrent VTE cohorts
Acute VTE events Figure 1 is a flow diagram showing inclusion and exclusion criteria for the index VTE cases. Potential cases were identified by finding the first hospital or emergency department diagnosis of VTE that occurred between 1 January 1997 and 2 October 2007. Codes that defined PE were: ICD-9-CM = 415.11, 415.19, 673.2, or 997.3 plus a PE code, or DRG = 078; codes that defined DVT were: ICD-9-CM = 453.2, 453.40, 453.41, 452.42, 453.8, 453.9, 671.33, 671.44, 671.9, or 997.2 plus a DVT code, or DRG = 128. In order to identify only cases with an incident VTE event, cases were excluded if there was: (i) a concurrent or previous code indicating a personal history of VTE (V12.51 or V12.52), (ii) a prior diagnosis of an acute venous thrombotic event (any ICD-9-CM 451, 453, 671 or 673.2 series code) back to 1 January 1991, or (iii) a prior placement of an inferior vena cava (IVC) filter (38.7). Surgery-provoked and unprovoked VTE cohorts To eliminate VTE cases associated with other major provoking risk factors, we used methods developed in past studies [21,22] to identify and then exclude incident VTE cases associated with cancer (concurrent or within prior 182 days), current pregnancy, recent major trauma (£ 90 days), concurrent or recent medical hospitalization (£ 60 days) or serious medical disease (systemic lupus, inflammatory bowel disease, sickle cell anemia, paraplegia etc., see Appendix). The remaining cases either had a principal diagnosis of VTE with no identifiable provoking risk factor (unprovoked VTE) or had recent major surgery (£ 60 days). If there was more than one ICD-9-CM VTE code listed (e.g. 415.11 and 453.41) a single code was assigned based on a pre-defined hierarchy: cases with PE (415.1x or 673.2 or Diagnostic Related Group-DRG 078) were selected first, then proximal DVT (453.41), then distal DVT (453.42), then DVT in the leg (453.40), then venous thrombosis of an Ôother specified veinÕ (453.8), then venous thrombosis of a Ônon-specified veinÕ (453.9), then DRG = 128 (DVT) and then pregnancy codes (671.33, 671.44 or 671.9). Each surgeryprovoked VTE case had to: (i) have a secondary code for VTE that was labeled as not being present on admission (POA = N, hospital acquired); cases indexed as POA = Y were excluded, or (ii) have a readmission to a hospital or an emergency department visit with a principal diagnosis of VTE within 60 days of the operation date. The date of diagnosis of all hospital VTE events was arbitrarily defined as the median hospital day. Outcome- recurrent VTE events Three criteria were used to define a recurrent VTE event: (i) re-hospitalization or an ED encounter one or more days after the index VTE encounter with a principal diagnosis of DVT or PE, using the same set of ICD-9-CM codes and DRG codes used to define index VTE events, (ii) presence of a highly specified ICD-9-CM code (415.11, 415.19, 453.40–453.42, 671.33) in a secondary position during a subsequent hospitalization only 2010 International Society on Thrombosis and Haemostasis
Recurrent venous thromboembolism 989 All cases with hospital discharge or emergency department ICD-9-CM diagnosis of acute VTE between Jan 1,1997-Oct 2, 2007 were identified. A specific ICD-9-CM code for VTE had to be present in the principal or secondary position for pulmonary embolism (415.11, 415.19) or deep-vein thrombosis (453.2, 453.40-453.42, 453.8, 453.9) or pregnancy thromboembolism (673.2, 671.33, 671.44, 671.90). Codes 997.2 and 997.3 were acceptable in the principal position if followed by one of the specified VTE codes. DRG 078 and DRG 128 were accepted. Excluded cases with prior VTE: a) = V12.51 or V12.52 or b) = any VTE code (451.x, 453.x, 671.x, 673.2) during any hospitalization back to 1/1/1991 or emergency department encounter back 1/1/05
Excluded cases with prior thrombotic events, post-thrombotic syndrome, pulmonary hypertension or placement of IVC filter = 459.1, 453.0, 453.3, 416.0, 416.8, 451.9, 453.1; procedure ICD-9CM = 38.7,back to 1/1/91 or CPT = 376.20 759.40, back to 1/1/05.
Excluded cases if the acute VTE was associate with: A) Active cancer (< 6 months) B) Pregnancy (ante-partum or < 6 weeks post-partum) C) Major trauma (≤ 90 days) D) Medical hospitalization < 60 days (except if they were admitted with a principal diagnosis of VTE) E) Serious medical illness (e.g. lupus, sickle cell, paralysis, inflammatory bowel disease etc) F) More than one major operation within prior 61 days
Excluded cases with secondary diagnosis of VTE during the hospitalization for the surgery if it was indexed as “condition present on admission” (POA = Y) or POA = unknown or POA = missing.
Include only cases that had a principal procedure code used to define one of 9 types of operations. and the operation had to be performed on hospital day 0 or 1. All VTE Cases had to meet one of the following criteria 0-60 days after the operation: A) Presence of a secondary code for VTE indexed as not being present on admission” (POA = N, hospital acquired), or B) readmitssion to a hospital or diagnosed in an emergency department with principal diagnosis of VTE within 60 days of the operation.
Fig. 1. Schematic diagram showing the inclusion and exclusion criteria for thomboembolism cohorts that were assembled.
if indexed as not being present on admission (POA = N), or (iii) presence of a less highly specified ICD-9-CM code for VTE (453.8 or 453.9) if indexed as POA = N and accompanied by a procedure code for a venogram (88.66), venous ultrasound imaging (88.77), injection of an anticoagulant (99.19), or interruption of the inferior vena cava (38.7). Each recurrent VTE event was classified in a hierarchical fashion based on the presence or absence of a specified risk factor in the following order : (i) active cancer, (ii) a recent delivery (< 6 weeks) or current pregnancy, (iii) recent major trauma (£ 90 days), (iv) the index VTE event diagnosed within 60 days, (v) recent surgery (surgery DRG) within 60 days, (vi) recent medical hospitalization in the prior 60 days, and (vii) presence of a serious medical condition. The remaining cases were classified as unprovoked or ÔidiopathicÕ recurrent VTE. 2010 International Society on Thrombosis and Haemostasis
Operations analyzed
We specifically identified cases coded as having a surgical DRG code plus one of the following principal procedure codes occurring on hospital day 0 or 1 (elective or emergent): incision or excision of brain or spinal cord: laminectomy, or excision of an intervertebral disc; coronary artery bypass graft surgery; colorectal resection; cholecystectomy with common duct exploration; hysterectomy (abdominal or vaginal); partial or total knee replacement; hip replacement (total or partial); and spinal fusion/repair surgery. Co-morbid medical conditions
The presence of co-morbid medical disorders was determined using a modification of the Healthcare Costs and Utilization
990 R. H. White et al
Project (or ÔElixhauserÕ) co-morbidity software [23]. Among the 29 chronic medical conditions included in this index, two conditions that reflect acute illness rather than chronic illness (fluid and electrolyte disorder, coagulation deficiency) were not included. Three other Elixhauser groups reflect the presence of cancer, and because these cases were already excluded, these cancers groups were not included (lymphoma, solid cancer and metastatic cancer), leaving terms for 24 chronic co-morbid conditions (e.g. chronic renal disease, hypertension, heart failure, depression etc.). Presence of any of these conditions for the index surgical hospitalization or any other hospitalizations in the prior 12 months was used to define the presence of comorbid conditions. Studies that have analyzed the validity of coded co-morbid conditions indicate very good agreement between chart review and the administrative data [24–26].
Among the selected cases, 29 138 (1.1%) met criteria for having acute VTE within 60 days of the operation. Of these, a further 13 884 (48%) cases were excluded because they had: active cancer (n = 6171), a concurrent pregnancy/puerperium (n = 63), recent major trauma (n = 1230), an intervening medical hospitalization within 60 days (n = 2406), a serious underlying medical illness (n = 4010), or underwent two major operations in the 60 days prior to the VTE (n = 78). In these later cases, the first of the two operations was excluded. An additional 3379 cases were excluded because the VTE ICD-9CM code was in a secondary diagnosis position with the present-on-admission index code equal to yes, unknown or missing. These last cases were not included because the predictive value of these codes was lower than 80%. Unadjusted characteristics of cohorts
Matching and statistics
The principal analysis was a matched comparison of the Kaplan–Meier incidence of recurrent VTE, which allowed a direct comparison of the observed incidence rates of recurrent VTE between groups. Cases undergoing each specified type of operation that were diagnosed as having acute VTE within 60 days of the date of surgery were matched with three cases with unprovoked VTE based on: age (± 2 years), race, gender, the ICD-9-CM VTE code (i.e. PE code, specific DVT code, non-specific DVT code etc.), the year of the index event and the number of co-morbidities. Exact matching was required for the VTE code and gender. Remaining matches were scored (15 point scale) for the precision of matching based on age, year of diagnosis, race and number of co-morbidities; 95% of all matches had a score of 14 or more and 90% were perfect matches. Cases undergoing each of the nine types of operation were matched with unprovoked cases without replacement; a separate match using all surgery cases combined was also generated. Cases could be followed until death or diagnosis of recurrent VTE. The hazard function at the mid-point for of the each specified time intervals was estimated using the actuarial method applied to the matched sets. Differences in hazard rates were compared using FiellerÕs method [27]. To compare the effect of gender on the incidence of recurrent VTE, cases within each surgery cohort were first stratified by gender before matching on the same parameters within the surgical group prior to matching with the unprovoked cases. Categorical data were analyzed using v2 testing. Analyses were performed using SAS and a two-sided P-value
