A total of 1080 records from the aforementioned databases and 45 records obtained by manual retrieval were collected. Among them, 251 duplicates were deleted, and 832 records were removed according to the inclusion and exclusion criteria after examining their titles and abstracts. When the full text was examined, a total of 22 studies were filtered out because they were inappropriate for inclusion in the meta-analysis (one study did not report any of the outcomes that we were interested in, 19 studies were case series without a control group, and two studies enrolled fewer than ten patients). Ten studies (4, 9, 14, 15, 21-26) were ultimately included in this meta-analysis and systematic review (Fig 1).
The included studies were assessed with the NOS scale, and the average score was 7.2 points, with no study scoring less than 6 points. Four studies (4, 9, 21, 25) did not clearly report the follow-up period. Six studies (14, 15, 21-23, 26) either did not report some important patient factors, such as age, or no solution was taken in their studies when there was a significant difference in demographics. Detailed scores for each study are presented in Supplementary 3.
Ten studies (4, 9, 14, 15, 21-26) with 6904 patients from four countries were included in this review, all of which were retrospective cohort studies published from 1995 to 2019. One study (23) included both total hip arthroplasties and hemiarthroplasties, and only the former were included in our meta-analysis. Total hip arthroplasties were studied in nine studies (4, 9, 14, 15, 21-24, 26), and total knee arthroplasties were studied in two studies (24, 25). Four studies (4, 9, 21, 25) collected data from national databases, while the other six studies (14, 15, 22-24, 26) used data from the authors' own institutions with mid-term to long-term follow-up (from 44 to 132 months). Four studies(21, 23, 24, 26) clearly expressed that only hemodialysis patients were included in the dialysis group, but the other six studies (4, 9, 14, 15, 22, 25) did not provide information on whether peritoneal dialysis or hemodialysis was applied. In four studies (14, 15, 24, 26), patients in the renal transplant group had a younger average age compared with the dialysis group. Similarly, in another two studies (9, 21), there was a lower percentage of older patients in the renal transplant group. (Table 1)
Primary outcomes
1. Mortality
Mortality was reported in six studies, but the integrated result showed high heterogeneity (Supplementary 4). Therefore, a sensitivity analysis was performed, and it was found that Cavanaugh's study (4) might be the potential source of the observed heterogeneity. The mortality reported in Cavanaugh's study only included inpatient deaths, while other studies included all deaths. Therefore, Cavanaugh's study was excluded from the meta-analysis of mortality. After excluding Cavanaugh's study, the results, of five studies (9, 14, 15, 22, 23) with a total of 505 patients, demonstrated a lower risk of mortality in the renal transplant group than in the dialysis group (RR=0.56, Cl= [0.42, 0.73], P<0.01) with moderate heterogeneity (I2=49%) (Fig 2).
2. Revision rate.
Data on revision were presented in nine included studies (9, 14, 15, 21-26) involving 4172 joints. A lower risk of revision was shown in the renal transplant group in the meta-analysis (RR =0.42, CI = [0.30, 0.59], P < 0.01), and the heterogeneity of the nine studies was acceptable (I2=43%) (Fig 3).
3. Periprosthetic joint infection
In nine studies, 445 of the total 4172 joints were infected, and the overall heterogeneity was high (I2=61%). A subgroup analysis was conducted to reduce the heterogeneity and explore the potential source of heterogeneity. Studies with sample sizes over larger than 100 were separated from those with sample sizes less than 100, and both subgroups had low heterogeneity. Six studies (14, 15, 22, 23, 25, 26) with 243 hips were included in the small-sample-size subgroup and no significant difference in risk of infection was detected between the renal transplant group and dialysis group (RR =0.83,CI =[0.40, 1.73], P= 0.62, I2=0%). In contrast, in the large-sample-size subgroup, which involved three studies (9, 21, 24) and 3929 joints, significantly lower risk of infection was shown in the renal transplant group (RR =0.19,CI =[0.13, 0.23], P < 0.01, I2=0%) (Fig 4).
Secondary outcomes
Overall complications were reported in five studies (9, 14, 23, 24, 26) with 604 joints. A random effect model was used to address the high heterogeneity, and the results revealed no significant difference in the risk of overall complications between the two groups (RR =0.72, CI = [0.50, 1.06], P = 0.13). Similarly, there was no difference in the rate of dislocation and or venous thrombosis between the two groups. (RR =1.29, CI = [0.45, 3.72], P = 0.63; RR =0.87, CI = [0.56, 1.35], P = 0.54, respectively) (Fig 5).
Publication bias
There was no apparent asymmetry in the funnel plot of revision rate, and it was inferred that a low risk of publication bias existed in those studies (Fig 6).