Data sources

We performed this registry-based historical cohort study in Denmark, which has 5.6 million inhabitants. All Danish residents are given universal, tax-paid access to hospitals. The Danish National Patient Registry (NPR) is a nationwide registry that has covered admissions to non-psychiatric hospitals since 1977 and outpatient and emergency room visits since 1995. The data include relevant dates and discharge diagnoses, coded in accordance with the International Classification of Diseases, 10th edition (ICD-10) from 1994 and the ICD-8 before that (Lynge et al. 2011). The Danish Hip Arthroplasty Registry (DHR) is a clinical database of all primary or revision total hip arthroplasties performed in Denmark since January 1, 1995. The data are entered by the operating surgeon immediately after the procedure and include the indication for arthroplasty (primary osteoarthritis, fracture, avascular necrosis, or other indication) (Pedersen et al. 2004). The indication for arthroplasty has been confirmed by medical chart review and radiographs in 79 of 80 randomly selected AVN patients (Pedersen et al. 2004). The Danish Central Office of Civil Registration continuously monitors the vital status of Danish residents, including dates of emigration or death, and it issues a unique personal identification number to everyone at birth or immigration. This number enables linkage of individual-level data between the NPR, the DHR, and the civil registration system (Pedersen et al. 2006).

Cirrhosis patients and reference individuals

We identified all Danish residents with a first-time hospital discharge diagnosis of alcoholic cirrhosis (ICD-10: K70.3, K70.4) or unspecified cirrhosis (ICD-10: K74.6) between 1994 and 2011. Biopsy or clinical evaluation had confirmed 85% of diagnoses for cirrhosis in the NPR in a previous validation study (Vestberg et al. 1997). We defined the “index date” as the date of the first cirrhosis diagnosis. To study the association between cirrhosis and a total hip arthroplasty for AVN, we excluded cirrhosis patients if they had a previous diagnosis of avascular necrosis (ICD-10: M87.0), if they had previously undergone total hip arthroplasty, or if they were diagnosed with hip fracture (ICD-8: 820.xx, 821.xx, 822.xx, 823-xx; ICD-10: S72.0, S82.0, S82.1, S83.x) before the index date. We matched these cirrhosis patients 1:5 on the basis of age, sex, and birth date to reference individuals without cirrhosis from the general Danish population, using risk-set sampling (Langholz and Goldstein 1996). The reference individuals were given the same index date as the corresponding cirrhosis patient. We excluded reference individuals according to the same criteria as cirrhosis patients; this exclusion resulted in a situation whereby not all cirrhosis patients were matched 1:5.

Confounders

The NPR holds data on potential confounders of an association between cirrhosis and AVN. We identified previous emergency room visits, inpatient and outpatient hospitalizations for conditions predisposing to AVN (diabetes, HIV infection, myeloproliferative disease, hemoglobinopathy, chronic renal failure, gout, and solid organ transplantation), an indicator of smoking (chronic obstructive pulmonary disease), and indicators of corticosteroid treatment (autoimmune hepatitis, rheumatoid arthritis, and connective tissue disease) (diagnosis codes are shown in Table 1). As an indicator of alcohol intake, we identified emergency room visits and inpatient and outpatient hospitalizations for alcoholism or alcohol-related disorders before the index date (see Supplementary data, Table 3).

Outcomes and statistical analysis

We examined one outcome: time to total hip arthroplasty for AVN. We followed the cirrhosis patients and the reference individuals from the index date to the date of total hip arthroplasty for AVN, date of death, or end of follow-up (December 31, 2011). We used stratified Cox regression to estimate the hazard ratio (HR) of total hip arthroplasty for AVN in cirrhosis patients as opposed to reference individuals and adjusted these HRs for potential confounders. We found no violations of the proportional hazards assumption when we tested it using Schoenfeld residuals and checked it by inspecting the log-log plot. We used the cumulative incidence function with death as a competing risk to compute the 5-year risk of total hip arthroplasty for AVN. This analysis relies on non-informative censoring. There were 2 censoring events in our study cohort: end of study (on December 31, 2011) and migration. Both of these events are unlikely predictors of the risk of a total hip arthroplasty for AVN, so the censoring in our study cohort was non-informative. Alcohol intake is a well-known risk factor for AVN, and cirrhosis patients have a high prevalence of alcohol intake. We were concerned that the regression analysis would leave residual confounding, so we performed a supplementary analysis in which we used restriction to minimize confounding by alcohol intake. We repeated the regression analysis and restricted it to cirrhosis patients with unspecified cirrhosis (ICD-10: K74.6) who had not been hospitalized for an alcohol-related disorder (see Supplementary data, Table 3), and the corresponding reference individuals. Reference individuals who had previously been hospitalized for an alcohol-related disorder were also left out of this analysis. All statistical analyses were performed using Stata version 12.1 and the R software package version 2.14 (R 2013).

Results

We included 25,421 cirrhosis patients and 114,052 reference individuals. Their median age was 57 years and 65% were male. 45 cirrhosis patients and 44 reference individuals underwent total hip arthroplasty for AVN. Diabetes and COPD were the most prevalent confounders, and the majority of confounders were more prevalent in cirrhosis patients than in reference individuals (Table 1). Cirrhosis patients’ adjusted HR for a total hip arthroplasty for AVN was 10 (95% CI: 6–17). Both cirrhosis patients’ and reference individuals’ 5-year risk of a total hip arthroplasty for AVN was very low, but it was markedly higher in cirrhosis patients: 0.16% (95% CI: 0.12–0.23) vs. 0.02% (95% CI: 0.01–0.03). Cirrhosis patients’ HR for a total hip arthroplasty for AVN was essentially unaltered in our supplementary analysis, restricted to patients without alcoholic cirrhosis (Table 2).

Supplementary data

Table 3 is available on the Acta Orthopaedica website (www.actaorthop.org), identification number 7581.

No competing interests declared.

TD and PJ analyzed and interpreted the data. TD, SO, HV, and PJ conceived and designed the study, and wrote the manuscript.