Journal of Thrombosis and Haemostasis, 13: 206–213
DOI: 10.1111/jth.12805
ORIGINAL ARTICLE
Risks of venous thromboembolism in patients with liver cirrhosis: a nationwide cohort study in Taiwan K . - J . N G , * † Y . - K . L E E , * ‡ M . - Y . H U A N G , § C . - Y . H S U ¶ and Y . - C . S U * ‡ *School of Medicine, Tzu Chi University, Hualien; †Department of Internal Medicine, Dalin Tzu Chi Hospital, Buddhist Tzu Chi Medical Foundation; ‡Emergency Department, Dalin Tzu Chi Hospital, Buddhist Tzu Chi Medical Foundation, Chiayi; §Department of Emergency Medicine, Mackay Memorial Hospital; and ¶Department of Public Heath, National Taiwan University, Taipei, Taiwan
To cite this article: Ng KJ, Lee YK, Huang MY, Hsu CY, Su YC. Risks of venous thromboembolism in patients with liver cirrhosis: a nationwide cohort study in Taiwan. J Thromb Haemost 2015; 13: 206–13.
Summary. Background: The results of various studies attempting to assess the risks of venous thromboembolism in liver cirrhosis have been conflicting. Furthermore, although the incidence of venous thromboembolism is thought to be low in Asians, the relationship between venous thromboembolism and liver cirrhosis has not been investigated in Asian countries. Objective: We investigated the risks of venous thromboembolism in cirrhotic patients in Taiwan to evaluate whether the risk is higher than in the general population. Methods: The data from 1 000 000 National Health Insurance beneficiaries were utilized. All adult beneficiaries were followed from 1 January 2007 to 31 December 2010 to identify those who developed venous thromboembolism. Each identified patient with liver cirrhosis was matched with 10 non-cirrhotic patients on the basis of high-dimensional propensity score. Cox regression models were applied to compare the hazards of venous thromboembolism in the matched cohorts. Results: A total of 757 940 patients were enrolled. After matching, 2223 cirrhotic patients and 22 230 non-cirrhotic patients were selected. The adjusted hazard ratio of venous thromboembolism was significantly increased by having cirrhosis (1.71; 95% confidence interval [CI] 1.05–2.78). A subgroup analysis revealed a much higher hazard ratio of venous thromboembolism in an advanced cirrhosis subgroup (n = 293) than in a matched non-cirrhosis subgroup (n = 2930) (4.36; 95% CI 1.36–14.01). Conclusion: The risk of venous thromboembolism may be higher in Asian patients with
Correspondence: Yung-Cheng Su, Emergency Department, Dalin Tzu Chi Hospital, Buddhist Tzu Chi Medical Foundation, No. 2, Minsheng Rd, Dalin Township, Chiayi County 622, Taiwan. Tel.: +886 5 2648000 ext 5838; fax: +886 5 2648499. E-mail: [email protected] Received 21 July 2014 Manuscript handled by: M. Cushman Final decision: F. R. Rosendaal, 27 November 2014
cirrhosis than in the general Asian population, especially in those with advanced cirrhosis. Keywords: cohort studies; liver cirrhosis; pulmonary embolism; venous thromboembolism; venous thrombosis.
Introduction Venous thromboembolism (VTE) is an important disease because it increases morbidity and often complicates the course of hospitalized patients [1,2]. During 2007–2009, a discharge diagnosis of VTE was recorded in an annual average of 547 596 adult hospitalizations in the USA [3]. Apart from VTE management, early identification is important to reduce morbidity and mortality. Many precipitating factors have been found to be associated with the occurrence of VTE, such as increasing age, malignancies, major surgery, and heart failure [1,2]. Another precipitating factor, liver cirrhosis (LC), has become an increasing concern [4–6]. Traditionally, bleeding has been regarded as the most frequent hemostatic complication of LC, and it is generally assumed that patients with advanced liver disease or cirrhosis are ‘auto-anticoagulated’. However, the protective effect of coagulation defects and hemostasis imbalance in cirrhotic patients has recently been called into question [4,6–8]. The results of various studies attempting to assess VTE risk in LC have been conflicting, with some studies showing that the risk of VTE is lower than in the general population [7,9,10], and others showing that it is higher [6,7,11]. Furthermore, although the incidence of VTE is thought to be low in Asians, partly because of the lower prevalence of predisposing genetic factors such as factor V Leiden [2,12], the relationship between VTE and LC has not been studied in Asian populations. In this study, a large administrative database was used to evaluate the risks of VTE in cirrhotic patients in Taiwan. Also, high-dimensional propensity score (hdPS) matching (i.e. a relatively new semi-automated statistical © 2014 International Society on Thrombosis and Haemostasis
Higher risk of venous thromboembolism in liver cirrhosis 207
method) was used to address possible confounding. The results of this study may help clinicians to identify individuals at risk for this frequently encountered disease. Methods Ethics statement
This study was initiated after its protocol had been approved by the Institutional Review Board of Dalin Tzu Chi Hospital, Buddhist Tzu Chi Medical Foundation, Taiwan, and was conducted in conformity with the Declaration of Helsinki. Database
The National Health Insurance (NHI) program was implemented in Taiwan in 1995, and provides compulsory universal health insurance. It enrolls ~ 99% of the Taiwanese population, and has contracts with 97% of all of the country’s medical providers [13,14]. The database contains comprehensive information on all insured subjects, including sex, date of birth, residential or work area, dates of clinical visits, diagnoses identified by International Classification of Diseases (Ninth Revision) Clinical Modification (ICD-9-CM) diagnostic codes, details of prescribed medications and received procedures, expenditure amounts, and outcome at hospital discharge (i.e. recovered, died, or transferred out). A random sample of 1 000 000 people receiving health benefits from the NHI program was selected on the basis of year 2005 reimbursement data, and was considered to be representative of the entire population; the group did not differ statistically from the larger cohort in age, sex, or healthcare costs, according to the Taiwan National Health Research Institute [15]. This sample was used as our study cohort. Study population
The population sample was followed from 1 January 2005 to 31 December 2010 (a total of 6 years). First, we identified those people who were still alive in 2007 and aged > 18 years as our study cohort, and selected patients with LC diagnoses (ICD-9-CM codes: 571.2, 571.5, and 571.6) from records of their ambulatory care claims between 2005 and 2006. To avoid coding error, an individual was classified as cirrhotic only if the individual had an LC diagnosis and then experienced another one or more diagnoses within the subsequent 12 months. Moreover, the first and last visits during the 2-year period had to be separated by at least 30 days to avoid accidental inclusion of patients with miscoded diagnoses. This selection process has been well validated and accepted [16]. To avoid creating financial hardship for patients with major illnesses, the NHI specifies 31 categories of catastrophic illness (e.g. dementia, cancers, and chronic renal © 2014 International Society on Thrombosis and Haemostasis
failure) that are exempt from copayment. The attending physician of a patient diagnosed as having a catastrophic illness can submit related information in an application for a catastrophic illness certificate. A committee formally reviews applications, and, if approved, patients are then exempted from copayment [17]. Those patients with advanced stages of LC would be registered. The criteria used by the NHI database to define advanced disease are: (i) intractable ascites; (ii) a history of hepatic encephalopathy; or (iii) a history of variceal-related hemorrhage. We also identified subgroups of patients in the LC group for further analysis. VTE was identified by ICD-9-CM codes for either deep vein thrombosis (451.1x, 451.81, 451.83, 453.x, 671.3x, and 671.4x) or pulmonary embolism (415.1x and 673.2x) in any clinical record [3]. To maximize case ascertainment, only patients hospitalized with VTE were included [18], and those diagnosed with VTE for any cause before 1 January 2007 were excluded. Each person was tracked for a 4-year period from 1 January 2007 to 31 December 2010 to establish the diagnosis of VTE. The administrative data of these patients for the period 2007–2010 were used to calculate disease-free survival time. Patients who either withdrew from the NHI program or remained healthy at the end of the follow-up period were censored (Fig. 1). Validation of diagnosis and outcome
We validated the ICD-9-CM code for the identification of LC and VTE by analyzing the medical records randomly selected from the electronic database in the year 2010 at Dalin Tzu Chi Hospital, Buddhist Tzu Chi Medical Foundation, a teaching hospital in southern Taiwan. LC and VTE were identified by imaging studies. Positive predictive value (PPV) was estimated. The diagnosis was confirmed by chart review in 47 of 50 randomly selected patients with codes for VTE, indicating a PPV of 94.0% (95% confidence interval [CI] 83.5–98.7%). Additionally, five patients developed VTE during hospitalization, and 42 patients were admitted mainly for management of VTE. Among 100 randomly selected patients with codes for LC, 98 were confirmed to have LC, indicating a PPV of 98% (95% CI 93.0–99.8%). Prespecified covariates
To better understand the effect of LC on the risk of VTE, several covariates were selected, including age, sex, urbanization level (i.e. urban, suburban, and rural), and socioeconomic status (SES). Income-related insurance payment amounts were used as a proxy measure of individual SES at follow-up. People were divided into three groups: (i) low SES, i.e. payment lower than US$571 per month (New Taiwan dollars [NT$] 20 000); (ii) moderate SES, i.e. payment of US$571–1141 per month (NT
208 K.-J. Ng et al
989 194 People were enrolled from 1 January 2005
Excluded 203 791 Aged < 18 years before 2007 540 Diagnosed with VTE before 2007 26 923 Dropped out before 2007
757 940 (aged ≥ 18 years) Were followed from 1 January 2007 to 31 December 2010
2779 People in the LC group
755 161 People in the unexposed group
2223 Were selected
437 Diagnosed as having LC with complications
1 : 10 Match
22 230 Were selected
1:10 Match
293 were selected
2930 were selected
Fig. 1. Flow diagram of the population-based study. LC, liver cirrhosis; VTE, venous thromboembolism.
$20 000–40 000); and (iii) high SES, i.e. payment of US $1142 or more per month (NT$40 001 or more) [17]. Then, the prevalence of selected comorbid conditions (i.e. diabetes, hypertension, coronary artery disease, hyperlipidemia, malignancies, congestive heart failure, atrial fibrillation, smoking, obesity, and peripheral artery disease) and the Charlson Comorbidity Index (CCI) score were determined by the use of discharge diagnoses either
during outpatient clinic visits or hospitalizations before 1 January 2007. The detailed ICD-9-CM codes of other comorbidities are listed in Table S1, and the processes for selecting comorbidities have been widely used and accepted [19]. The CCI is a scoring system that assigns weights to important concomitant diseases; it has been validated for use in studies employing ICD-9-CM data [20]. © 2014 International Society on Thrombosis and Haemostasis
Higher risk of venous thromboembolism in liver cirrhosis 209
The hdPS algorithm is available as an SAS macro from the Brigham and Women’s Hospital [21], and the details are described elsewhere [22]. Briefly, it is an empirically driven, multistep algorithm used to adjust for confounding bias. The macro can identify candidate covariates from predefined data dimensions, which, in this study, are the procedures received and diagnoses (codes) reported between 1 January 2005 and 31 December 2006. Next, it automatically assesses recurrence (i.e. repetition of the same code), prioritizes covariates, and selects covariates for adjustment. In this study, the 500 variables most likely to cause bias were selected. Together with the prespecified covariates, these 500 variables are included in a logistic regression model to estimate the predicted probability (the propensity score) of exposure to LC vs. the general population, conditional on all of the included covariates. We then matched each patient in the LC group to 10 patients in the non-LC group with the closest propensity scores, by using a standard greedy-matching algorithm [23,24], and compared the risk of VTE between these groups.
lege Station, TX, USA) were both used for data analysis. All covariates were taken as categorical variables, except for age and propensity score, which were treated as continuous variables. Categorical variables were compared by use of Pearson’s chi-square test (and continuous variables by use of the t-test) to determine baseline heterogeneity in the two groups. Kaplan–Meier curves were first plotted to show the cumulative risks of VTE. Cox proportional hazard regression models were then used to calculate the hazard ratios (HRs) of VTE for people with LC in the matched group after adjustment for age, sex, urbanization level, SES, diabetes, hypertension, coronary artery disease, hyperlipidemia, malignancies, congestive heart failure, atrial fibrillation, smoking, obesity, peripheral artery disease, and CCI score. A subgroup analysis was performed to evaluate the risk of VTE in patients with advanced-stage cirrhosis, to determine whether the relationship between LC and VTE was dependent on LC severity. These patients with advancedstage LC were also matched to 10 patients in the non-LC group on the basis of propensity score, and the same analyses were carried out on these subgroups. A two-tailed P-value of < 0.05 was considered to be significant.
Statistical analysis
Results
The SAS statistical package, version 9.3 (SAS Institute, Cary, NC, USA) and STATA version 11.2 (StataCorp, Col-
The distribution of demographic characteristics and selected morbidities in both groups is shown in Table 1.
hdPS
Table 1 Baseline characteristics of the liver cirrhosis (LC) group and the non-LC group
Variables Male, no. (%) Mean age in years (SD) Socioeconomic status, no. (%) Low Moderate High Urbanization level, no. (%) Urban Suburban Rural Diabetes, no. (%) Hypertension, no. (%) Coronary artery disease, no. (%) Hyperlipidemia, no. (%) Malignancies, no. (%) Heart failure, no. (%) Atrial fibrillation, no. (%) Charlson Comorbidity Index score, no. (%) 0 1 ≥2 Smoking, no. (%) Obesity, no. (%) Peripheral artery disease, no. (%) Mean propensity score (SD)
LC group (n = 2779)
Control group (n = 755 161)
1836 (66.1) 59.0 (13.4)
366 742 (48.6) 43.9 (16.7)
1476 (53.1) 1066 (38.4) 237 (8.5)
325 530 (43.1) 304 356 (40.3) 125 275 (16.6)
622 1252 905 805 1060 655 470 793 161 43
(22.4) (45.0) (32.6) (29.0) (38.1) (23.6) (16.9) (28.5) (5.8) (1.6)
229 771 345 063 180 327 57 401 124 366 86 703 76 606 23 056 9762 3260
(30.4) (45.7) (23.9) (7.6) (16.5) (11.5) (10.1) (3.1) (1.3) (0.4)
(0) (18.8) (81.2) (1.2) (0.4) (2.2) (0.3424)
507 055 149 465 98 641 5773 2493 4525 0.0058
(67.1) (19.8) (13.1) (0.8) (0.3) (0.6) (0.0189)
0 523 2256 32 12 60 0.1847
SD, standard deviation. P-Values less than 0.05 are presented as bold types. © 2014 International Society on Thrombosis and Haemostasis
P-value < < < – – – < – – – < < < < < < <
DOI: 10.1111/jth.12805
ORIGINAL ARTICLE
Risks of venous thromboembolism in patients with liver cirrhosis: a nationwide cohort study in Taiwan K . - J . N G , * † Y . - K . L E E , * ‡ M . - Y . H U A N G , § C . - Y . H S U ¶ and Y . - C . S U * ‡ *School of Medicine, Tzu Chi University, Hualien; †Department of Internal Medicine, Dalin Tzu Chi Hospital, Buddhist Tzu Chi Medical Foundation; ‡Emergency Department, Dalin Tzu Chi Hospital, Buddhist Tzu Chi Medical Foundation, Chiayi; §Department of Emergency Medicine, Mackay Memorial Hospital; and ¶Department of Public Heath, National Taiwan University, Taipei, Taiwan
To cite this article: Ng KJ, Lee YK, Huang MY, Hsu CY, Su YC. Risks of venous thromboembolism in patients with liver cirrhosis: a nationwide cohort study in Taiwan. J Thromb Haemost 2015; 13: 206–13.
Summary. Background: The results of various studies attempting to assess the risks of venous thromboembolism in liver cirrhosis have been conflicting. Furthermore, although the incidence of venous thromboembolism is thought to be low in Asians, the relationship between venous thromboembolism and liver cirrhosis has not been investigated in Asian countries. Objective: We investigated the risks of venous thromboembolism in cirrhotic patients in Taiwan to evaluate whether the risk is higher than in the general population. Methods: The data from 1 000 000 National Health Insurance beneficiaries were utilized. All adult beneficiaries were followed from 1 January 2007 to 31 December 2010 to identify those who developed venous thromboembolism. Each identified patient with liver cirrhosis was matched with 10 non-cirrhotic patients on the basis of high-dimensional propensity score. Cox regression models were applied to compare the hazards of venous thromboembolism in the matched cohorts. Results: A total of 757 940 patients were enrolled. After matching, 2223 cirrhotic patients and 22 230 non-cirrhotic patients were selected. The adjusted hazard ratio of venous thromboembolism was significantly increased by having cirrhosis (1.71; 95% confidence interval [CI] 1.05–2.78). A subgroup analysis revealed a much higher hazard ratio of venous thromboembolism in an advanced cirrhosis subgroup (n = 293) than in a matched non-cirrhosis subgroup (n = 2930) (4.36; 95% CI 1.36–14.01). Conclusion: The risk of venous thromboembolism may be higher in Asian patients with
Correspondence: Yung-Cheng Su, Emergency Department, Dalin Tzu Chi Hospital, Buddhist Tzu Chi Medical Foundation, No. 2, Minsheng Rd, Dalin Township, Chiayi County 622, Taiwan. Tel.: +886 5 2648000 ext 5838; fax: +886 5 2648499. E-mail: [email protected] Received 21 July 2014 Manuscript handled by: M. Cushman Final decision: F. R. Rosendaal, 27 November 2014
cirrhosis than in the general Asian population, especially in those with advanced cirrhosis. Keywords: cohort studies; liver cirrhosis; pulmonary embolism; venous thromboembolism; venous thrombosis.
Introduction Venous thromboembolism (VTE) is an important disease because it increases morbidity and often complicates the course of hospitalized patients [1,2]. During 2007–2009, a discharge diagnosis of VTE was recorded in an annual average of 547 596 adult hospitalizations in the USA [3]. Apart from VTE management, early identification is important to reduce morbidity and mortality. Many precipitating factors have been found to be associated with the occurrence of VTE, such as increasing age, malignancies, major surgery, and heart failure [1,2]. Another precipitating factor, liver cirrhosis (LC), has become an increasing concern [4–6]. Traditionally, bleeding has been regarded as the most frequent hemostatic complication of LC, and it is generally assumed that patients with advanced liver disease or cirrhosis are ‘auto-anticoagulated’. However, the protective effect of coagulation defects and hemostasis imbalance in cirrhotic patients has recently been called into question [4,6–8]. The results of various studies attempting to assess VTE risk in LC have been conflicting, with some studies showing that the risk of VTE is lower than in the general population [7,9,10], and others showing that it is higher [6,7,11]. Furthermore, although the incidence of VTE is thought to be low in Asians, partly because of the lower prevalence of predisposing genetic factors such as factor V Leiden [2,12], the relationship between VTE and LC has not been studied in Asian populations. In this study, a large administrative database was used to evaluate the risks of VTE in cirrhotic patients in Taiwan. Also, high-dimensional propensity score (hdPS) matching (i.e. a relatively new semi-automated statistical © 2014 International Society on Thrombosis and Haemostasis
Higher risk of venous thromboembolism in liver cirrhosis 207
method) was used to address possible confounding. The results of this study may help clinicians to identify individuals at risk for this frequently encountered disease. Methods Ethics statement
This study was initiated after its protocol had been approved by the Institutional Review Board of Dalin Tzu Chi Hospital, Buddhist Tzu Chi Medical Foundation, Taiwan, and was conducted in conformity with the Declaration of Helsinki. Database
The National Health Insurance (NHI) program was implemented in Taiwan in 1995, and provides compulsory universal health insurance. It enrolls ~ 99% of the Taiwanese population, and has contracts with 97% of all of the country’s medical providers [13,14]. The database contains comprehensive information on all insured subjects, including sex, date of birth, residential or work area, dates of clinical visits, diagnoses identified by International Classification of Diseases (Ninth Revision) Clinical Modification (ICD-9-CM) diagnostic codes, details of prescribed medications and received procedures, expenditure amounts, and outcome at hospital discharge (i.e. recovered, died, or transferred out). A random sample of 1 000 000 people receiving health benefits from the NHI program was selected on the basis of year 2005 reimbursement data, and was considered to be representative of the entire population; the group did not differ statistically from the larger cohort in age, sex, or healthcare costs, according to the Taiwan National Health Research Institute [15]. This sample was used as our study cohort. Study population
The population sample was followed from 1 January 2005 to 31 December 2010 (a total of 6 years). First, we identified those people who were still alive in 2007 and aged > 18 years as our study cohort, and selected patients with LC diagnoses (ICD-9-CM codes: 571.2, 571.5, and 571.6) from records of their ambulatory care claims between 2005 and 2006. To avoid coding error, an individual was classified as cirrhotic only if the individual had an LC diagnosis and then experienced another one or more diagnoses within the subsequent 12 months. Moreover, the first and last visits during the 2-year period had to be separated by at least 30 days to avoid accidental inclusion of patients with miscoded diagnoses. This selection process has been well validated and accepted [16]. To avoid creating financial hardship for patients with major illnesses, the NHI specifies 31 categories of catastrophic illness (e.g. dementia, cancers, and chronic renal © 2014 International Society on Thrombosis and Haemostasis
failure) that are exempt from copayment. The attending physician of a patient diagnosed as having a catastrophic illness can submit related information in an application for a catastrophic illness certificate. A committee formally reviews applications, and, if approved, patients are then exempted from copayment [17]. Those patients with advanced stages of LC would be registered. The criteria used by the NHI database to define advanced disease are: (i) intractable ascites; (ii) a history of hepatic encephalopathy; or (iii) a history of variceal-related hemorrhage. We also identified subgroups of patients in the LC group for further analysis. VTE was identified by ICD-9-CM codes for either deep vein thrombosis (451.1x, 451.81, 451.83, 453.x, 671.3x, and 671.4x) or pulmonary embolism (415.1x and 673.2x) in any clinical record [3]. To maximize case ascertainment, only patients hospitalized with VTE were included [18], and those diagnosed with VTE for any cause before 1 January 2007 were excluded. Each person was tracked for a 4-year period from 1 January 2007 to 31 December 2010 to establish the diagnosis of VTE. The administrative data of these patients for the period 2007–2010 were used to calculate disease-free survival time. Patients who either withdrew from the NHI program or remained healthy at the end of the follow-up period were censored (Fig. 1). Validation of diagnosis and outcome
We validated the ICD-9-CM code for the identification of LC and VTE by analyzing the medical records randomly selected from the electronic database in the year 2010 at Dalin Tzu Chi Hospital, Buddhist Tzu Chi Medical Foundation, a teaching hospital in southern Taiwan. LC and VTE were identified by imaging studies. Positive predictive value (PPV) was estimated. The diagnosis was confirmed by chart review in 47 of 50 randomly selected patients with codes for VTE, indicating a PPV of 94.0% (95% confidence interval [CI] 83.5–98.7%). Additionally, five patients developed VTE during hospitalization, and 42 patients were admitted mainly for management of VTE. Among 100 randomly selected patients with codes for LC, 98 were confirmed to have LC, indicating a PPV of 98% (95% CI 93.0–99.8%). Prespecified covariates
To better understand the effect of LC on the risk of VTE, several covariates were selected, including age, sex, urbanization level (i.e. urban, suburban, and rural), and socioeconomic status (SES). Income-related insurance payment amounts were used as a proxy measure of individual SES at follow-up. People were divided into three groups: (i) low SES, i.e. payment lower than US$571 per month (New Taiwan dollars [NT$] 20 000); (ii) moderate SES, i.e. payment of US$571–1141 per month (NT
208 K.-J. Ng et al
989 194 People were enrolled from 1 January 2005
Excluded 203 791 Aged < 18 years before 2007 540 Diagnosed with VTE before 2007 26 923 Dropped out before 2007
757 940 (aged ≥ 18 years) Were followed from 1 January 2007 to 31 December 2010
2779 People in the LC group
755 161 People in the unexposed group
2223 Were selected
437 Diagnosed as having LC with complications
1 : 10 Match
22 230 Were selected
1:10 Match
293 were selected
2930 were selected
Fig. 1. Flow diagram of the population-based study. LC, liver cirrhosis; VTE, venous thromboembolism.
$20 000–40 000); and (iii) high SES, i.e. payment of US $1142 or more per month (NT$40 001 or more) [17]. Then, the prevalence of selected comorbid conditions (i.e. diabetes, hypertension, coronary artery disease, hyperlipidemia, malignancies, congestive heart failure, atrial fibrillation, smoking, obesity, and peripheral artery disease) and the Charlson Comorbidity Index (CCI) score were determined by the use of discharge diagnoses either
during outpatient clinic visits or hospitalizations before 1 January 2007. The detailed ICD-9-CM codes of other comorbidities are listed in Table S1, and the processes for selecting comorbidities have been widely used and accepted [19]. The CCI is a scoring system that assigns weights to important concomitant diseases; it has been validated for use in studies employing ICD-9-CM data [20]. © 2014 International Society on Thrombosis and Haemostasis
Higher risk of venous thromboembolism in liver cirrhosis 209
The hdPS algorithm is available as an SAS macro from the Brigham and Women’s Hospital [21], and the details are described elsewhere [22]. Briefly, it is an empirically driven, multistep algorithm used to adjust for confounding bias. The macro can identify candidate covariates from predefined data dimensions, which, in this study, are the procedures received and diagnoses (codes) reported between 1 January 2005 and 31 December 2006. Next, it automatically assesses recurrence (i.e. repetition of the same code), prioritizes covariates, and selects covariates for adjustment. In this study, the 500 variables most likely to cause bias were selected. Together with the prespecified covariates, these 500 variables are included in a logistic regression model to estimate the predicted probability (the propensity score) of exposure to LC vs. the general population, conditional on all of the included covariates. We then matched each patient in the LC group to 10 patients in the non-LC group with the closest propensity scores, by using a standard greedy-matching algorithm [23,24], and compared the risk of VTE between these groups.
lege Station, TX, USA) were both used for data analysis. All covariates were taken as categorical variables, except for age and propensity score, which were treated as continuous variables. Categorical variables were compared by use of Pearson’s chi-square test (and continuous variables by use of the t-test) to determine baseline heterogeneity in the two groups. Kaplan–Meier curves were first plotted to show the cumulative risks of VTE. Cox proportional hazard regression models were then used to calculate the hazard ratios (HRs) of VTE for people with LC in the matched group after adjustment for age, sex, urbanization level, SES, diabetes, hypertension, coronary artery disease, hyperlipidemia, malignancies, congestive heart failure, atrial fibrillation, smoking, obesity, peripheral artery disease, and CCI score. A subgroup analysis was performed to evaluate the risk of VTE in patients with advanced-stage cirrhosis, to determine whether the relationship between LC and VTE was dependent on LC severity. These patients with advancedstage LC were also matched to 10 patients in the non-LC group on the basis of propensity score, and the same analyses were carried out on these subgroups. A two-tailed P-value of < 0.05 was considered to be significant.
Statistical analysis
Results
The SAS statistical package, version 9.3 (SAS Institute, Cary, NC, USA) and STATA version 11.2 (StataCorp, Col-
The distribution of demographic characteristics and selected morbidities in both groups is shown in Table 1.
hdPS
Table 1 Baseline characteristics of the liver cirrhosis (LC) group and the non-LC group
Variables Male, no. (%) Mean age in years (SD) Socioeconomic status, no. (%) Low Moderate High Urbanization level, no. (%) Urban Suburban Rural Diabetes, no. (%) Hypertension, no. (%) Coronary artery disease, no. (%) Hyperlipidemia, no. (%) Malignancies, no. (%) Heart failure, no. (%) Atrial fibrillation, no. (%) Charlson Comorbidity Index score, no. (%) 0 1 ≥2 Smoking, no. (%) Obesity, no. (%) Peripheral artery disease, no. (%) Mean propensity score (SD)
LC group (n = 2779)
Control group (n = 755 161)
1836 (66.1) 59.0 (13.4)
366 742 (48.6) 43.9 (16.7)
1476 (53.1) 1066 (38.4) 237 (8.5)
325 530 (43.1) 304 356 (40.3) 125 275 (16.6)
622 1252 905 805 1060 655 470 793 161 43
(22.4) (45.0) (32.6) (29.0) (38.1) (23.6) (16.9) (28.5) (5.8) (1.6)
229 771 345 063 180 327 57 401 124 366 86 703 76 606 23 056 9762 3260
(30.4) (45.7) (23.9) (7.6) (16.5) (11.5) (10.1) (3.1) (1.3) (0.4)
(0) (18.8) (81.2) (1.2) (0.4) (2.2) (0.3424)
507 055 149 465 98 641 5773 2493 4525 0.0058
(67.1) (19.8) (13.1) (0.8) (0.3) (0.6) (0.0189)
0 523 2256 32 12 60 0.1847
SD, standard deviation. P-Values less than 0.05 are presented as bold types. © 2014 International Society on Thrombosis and Haemostasis
P-value < < < – – – < – – – < < < < < < <
