Total knee arthroplasty using patient-specific instrumentation for osteoarthritis of the knee: a meta-analysis

We included studies which compared TKA using PSI (TKA-PSI) and standard TKA in this review. Standard TKA was defined as TKA using intramedullary alignment guiding nail for cutting femur and either intramedullary or extramedullary alignment guiding nail for cutting tibia. Studies were included if patients had primary TKA for knee OA classified into grade 3 and 4, according to Kellgren-Lawrence grading system [27]. If patients had rheumatic diseases or if less than 80% of the study’s population had OA, the studies were excluded. Since rheumatic diseases are systematic diseases, controlling for these conditions using medications directly influences clinical outcomes. Types of primary outcome measures are shown below:

As a secondary outcome measure, we investigated the percentage of alignment outliers from the planned alignment in the included studies in which PROMs were examined.

For patient-reported outcome measures (PROM), surgery time, blood loss, and transfusion rate, we included randomized controlled trials (RCT). For complications (i.e. SSI, DVT, revision TKA), we also included non-randomized comparative studies (non-RCT).

We conducted a comprehensive literature search for all relevant articles using four electronic databases: the Cochrane Central Register of Controlled Trials (CENTRAL) (The Cochrane Library 2019, Issue 2), MEDLINE (1946 to February 15th, 2019), and EMBASE (1974 to February 15th, 2019). We also examined ongoing trials using the database of clinical trials (https://​clinicaltrials.​gov). Each searching strategies are shown in the Additional file 1: Appendix B. In hand-searching, we screened the reference list of previous SRs and relevant studies for additional articles potentially not identified through electronic search. We chose the key search terms “knee*,” “arthroplasty OR replacement,” and “patient-specific OR patient-matched OR custom-fit OR custom-made OR custom*.” The search was limited to articles published since 2001, because PSI was initially used in 2001 in institutional studies only, and the first report using PSI was published in 2004 [28].

We adopted a 3-step screening process (title screening, abstract screening, and full-text screening) to select eligible articles. After duplicate articles were removed, two reviewers (KK AND AS) independently performed title and abstract screenings. If either of the reviewers included an article during title or abstract screening, it was moved to the next stage for screening. During full-text screening, discrepancies were resolved through discussion and consensus with the senior authors (FY AND OA). We did not register this protocol in time before data collection was performed. For each study, two reviewers independently extracted data into a spreadsheet for the outcomes designed a priori. Differences were resolved by discussion.

Two reviewers independently assessed risk of bias for RCTs using the Cochrane risk of bias tool [29]. The following domains were assessed: sequence generation; allocation concealment; blinding of participants and personnel (performance bias); blinding of outcome assessment (separately for PROM, transfusion, blood loss, surgery time, and complications); incomplete outcome data (attrition bias); and selective reporting (reporting bias). For non-RCTs included for complications, we used the methodological index for non-randomized studies (MINORS) appraisal tool [30].

Dichotomous outcomes (transfusion and complications) were expressed as a risk ratio (RR) with 95% confidence intervals (CI). Continuous outcomes (i.e. PROM, blood loss, and surgery time) were expressed as a mean difference (MD) between TKA-PSI and standard TKA groups. We preferred to calculate effect size measures (standardized mean difference (SMD)), when studies used different instruments to assess the same outcome (e.g. blood loss).

We analyzed outcomes according to the modified intention-to-treat method without imputation. When data was not expressed with mean and standard deviation (SD), but expressed with median, minimum-maximum range, or interquartile range, we estimated the mean and SD in reference to Wan et al. (2014) [31].

Heterogeneity between pooled studies were assessed using the chi-square test with statistical significance set at p < 0.10 and the I² statistic [32]. The I² value was assessed as follows: 0–40% might not be important, 30–60% moderate, 50–90% substantial, and 75–100% considerable. Also, we considered variance of the point estimate and overlap in the confidence intervals.

For assessing reporting biases, we constructed funnel plots for each outcome for which there were at least five trials. We pooled included studies using the generic inverse variance method for continuous outcomes and the Mantel-Haenszel method for dichotomous outcomes in Review Manager 5 (Revman). Either one of fixed-effect or random- effect model was used depending on the heterogeneity. We created a ‘Summary of findings’ for the main comparison. For PROM, we selected an outcome measure in which most patients were pooled each for less than 1-year and for 1-year or more. We pre-specified and carried out sensitivity analyses by excluding non-RCTs in complications. We assessed the quality of evidence related to each outcome measure using the Grading of Recommendations Assessment, Development and Evaluation (GRADE) approach [33].

We searched from February 15th 2019 to March 15th 2019. We screened a total of 1386 articles (CENTRAL 128, EMBASE 651, MEDLINE 607). After the removal of duplicates, we checked 913 articles in title-screening and 320 articles in abstract-screening processes. We added 13 articles from the references in the pooled articles and implemented full-text screening process. From the database of clinical trials, we included 1 study. Overall, 38 articles were included in this systematic review: 38 studies were included in qualitative synthesis and 37 studies were included in quantitative synthesis (meta-analysis), as seen in Fig. 1. Details of the process of screening are illustrated in PRISMA (Preferred reporting items for systematic reviews and meta-analyses) flowchart in Fig. 1.

We included 38 studies, with a total of 3487 patients (1753 in TKA-PSI, and 1734 in standard TKA) [7, 34‐71]. All studies except NCT02539992 were published in 2012–2018. We included 15 studies in PROM (14 studies in meta-analysis), 18 studies in surgery time (18 studies in meta-analysis), 15 studies in blood loss (15 studies in meta-analysis), 9 studies in transfusion rate (8 studies in meta-analysis), and 24 studies in complications (23 studies in meta-analysis). Boonen et al. (2013) and Boonen et al. (2016) were the same cohort and we counted them as the same study. Details of the included studies are provided in the characteristics of included studies table (Table 1).

In the included studies, 29 studies with pooled 2547 patients explicitly included patients with osteoarthritis, whereas 7 studies with pooled 726 patients recruited patients with knee deformity. Twenty-seven studies used MRI-based PSI, whereas 10 studies used CT-based PSI. Many varieties of PSI device were used and the most prominent device was Visionnaire (Smith & Nephew, Memphis, USA) in 15 studies followed by Materialise (Zimmer via Materialise, Belgium) in 10 studies. Thirteen studies with the pooled 1384 patients followed for 1-year or more.

In full-text screening process, we excluded 113 studies. We described studies in which discrepancy for agreement was found or excluded with special reasons, being listed in Additional file 1: Appendix C. We found 18 ongoing trials using the database of clinicaltrials.gov, and we searched their results using corresponding author’s name and institution where trials were conducted. A list of ongoing trials was shown in Additional file 1: Appendix D.

We assessed the risk of bias in the included studies. In sequence generation, 8 out of 25 RCTs properly generated unpredictable randomization lists, creating low risk. In allocation concealment, 9 studies appropriately distanced from recruiters notifying allocation, whereas two studies were exposed to high risk of selection bias by recruiting participants in turns between PSI and standard TKA groups [35], or using block randomization without blinding block size for recruiters [65]. PSI requires preoperative MRI or CT to create patient-specific cutting blocks, and we assumed patients were notified as to whether they were allocated to the TKA-PSI group. For that reason, in almost all RCTs except for two studies, there was a high risk of bias in the domain of blinding of outcome assessment for PROM. In the two studies, all patients had preoperative MRI or CT and blinding of patients was robustly maintained with low risk of bias [42, 70]. Also, due to the trait of surgical trials, surgeons and scrub nurses could not be blinded. The unblinding potentially caused performance bias for surgery time, blood loss, and transfusion rate with high risk of bias. In incomplete outcome data, the following outcomes (i.e. surgery time, blood loss, transfusion rate) were less likely influenced, since these were perioperatively recorded in medical charts during admission. Six studies in PROM had high risk of attrition bias [42, 51, 53, 58, 68, 71]. In the included studies for the percentage of outliers in the positioning of the prosthetic implants, five studies clearly performed blinding to the outcome assessors for the images, having low risk of outcome detection bias [45, 53, 57, 58, 67].

In non-RCTs, the mean MINORS scores were 16.8 (SD 3.9). Seven studies were prospective non-RCTs, whereas 6 studies were retrospective non-RCTs. In only one study, the main study purpose was measuring complications. We presented the MINORS scores in the included non-RCTs in Additional file 1: Appendix E.

The following outcome measures included two or more studies: KSS knee, KSS function, KSS total, Oxford, WOMAC, KOOS symptom, KOOS pain, KOOS ADL, KOOS sports, KOOS QoL, EQ-5D VAS, SF-12 physical score, and SF-12 mental score.

In patients followed for less than 1-year, 17 studies were included with the pooled 1609 patients. There were no significant differences in KSS knee, KSS function, KSS total, WOMAC, and Oxford scores between TKA-PSI and standard TKA (MD 0.24 (95%CI -2.25 – 3.65) in KSS knee at mean 3-months follow-up; MD 0.56 (95%CI -1.98 – 3.10) in KSS function at mean 3-months follow-up: MD -2.48 (95%CI -10.24 – 5.29) in KSS total at mean 3-months follow-up; MD 0.05 (95%CI -1.69 – 1.80) in WOMAC at mean 3-months follow-up, and MD -0.48 (95%CI -1.92 – 0.97) in Oxford score at mean 3-months follow-up). The MDs were less than minimally clinically important differences (MCID) in KSS knee (MCID 5.3), KSS function (MCID 6.1), WOMAC (MCID 10), and Oxford scores (MCID 4.3) [72‐74]. There were no significant differences in EQ-5D VAS, SF-12 physical and mental scores between groups: MD -0.56 (95%CI -6.57 – 5.45) in EQ-5D VAS at mean 3-months follow-up; MD 0.51 (95%CI -2.58 – 3.59) in SF-12 physical score at mean 3-months follow-up (MCID 4.5); MD 1.84 (95%CI -3.02 – 6.69) in SF-12 mental score at mean 3-months follow-up (MCID 3.3) [72].

In patients followed for 1-year or more, 13 studies were included with the pooled 1384 patients. There were no significant differences in Oxford, WOMAC, KSS knee and total, KOOS ADL and sports, and EQ-5D VAS between groups: MD 1.00 (95%CI -11.54 – 0.59) in KSS knee at 12-months follow-up, MD -5.22 (95%CI -10.70 – 0.06) in KSS function at 12-months follow-up, MD -2.51 (95%CI -14.18 – 9.15) in KSS total at mean 18.8-months follow-up, MD 2.66 (95%CI -1.34 – 6.67) in Oxford at mean 20.0-months follow-up, MD 0.25 (95%CI -4.39 – 4.89) in WOMAC at mean 21.6-months follow-up, MD 6.09 (95%CI 6.09–0.03 – 12.21) in KOOS ADL at mean 18.3-months follow-up, MD -2.39 (95%CI -19.99 – 15.20) in KOOS sports at mean 18.3-months follow-up, and MD -1.38 (95%CI -6.87 – 4.10) in EQ-5D VAS at mean 24-months follow-up). We found differences in KSS function, and KOOS symptom, pain, and QoL between groups, however, the pooled patients were small (less than 150) and the differences were less than the MCID: MD -5.34 (95CI -10.50 – − 0.18) in KSS function at mean 12-months follow-up (MCID 6.1), MD 5.23 (95%CI 0.11–10.35) in KOOS symptom at mean 18.3-months follow-up (MCID 10.7), MD 9.67 (95%CI 3.88–15.46) in KOOS pain at mean 18.3-months follow-up (MCID 16.7), MD 9.77 (95%CI 2.56–16.97) in KOOS QoL at mean 18.3-months follow-up (MCID 15.6) [73, 75]. We present more details in the forest plots in Figs. 2, 3, 4, 5, 6 and 7.

Two studies adopted unique outcome measures (i.e. Kujala [65] and Lysholm [35] scores), but they did not show differences in the outcome measures between TKA-PSI and standard TKA groups.

Eighteen studies were included with the pooled 1592 patients. There was high heterogeneity between studies with I² = 94%. The pooled mean difference was MD − 3.09 min (95%CI -6.73 – 0.55) and there was no significant difference between groups, as shown in Fig. 8.

Fifteen studies were included with various measuring scales and different time points when blood loss was measured postoperatively. In measuring blood loss, 5 studies measured volume of drainage fluid from suction drain device [34, 36, 39, 51, 67], 5 studies measured perioperative hemoglobin reduction [35, 42, 54, 64, 70], 3 studies measured intraoperative bleeding [7, 55, 68], 1 study calculated blood loss using the Mercuriali & Inghilleri formula [53, 76], and 1 study lacked information about the measurement [44]. The time points were most prominently at intraoperative in 3 studies, postoperative 24 h in 3 studies, 48 h in 2 studies, lowest hemoglobin in 1 study, and not shown in 4 studies. TKA-PSI decreased blood loss as compared to standard TKA with SMD -0.36 (95%CI -0.57 – − 0.15, p = 0.001) as seen in Fig. 9, although the effect size was small equivalent to hemoglobin 0.4 g/dl (95%CI 0.18–0.88) reduction. The effect size was estimated using the calculated SMD (− 0.36, 95%CI -0.57 - -0.15) and the pooled SD (1.2 g/dl) in hemoglobin decrease from a previous review examined blood loss after TKA [77, 78]: 0.43 g/dl (95%CI 0.18–0.88) = 0.36 (95%CI 0.15–0.57) × 1.2 g/dl.

Nine studies were chosen in qualitative synthesis and 8 studies were meta-analyzed with the pooled 487 patients. The overall transfusion rate was 17.3% (98/567), whereas two studies did not have any cases with transfusion. There was not significant difference in transfusion rate between TKA-PSI and standard TKA group with risk difference − 0.14 (95%CI -0.33 – 0.05, p = 0.16), as seen in Fig. 10. Pietsch et al. (2013) did not specify transfusion rate, but they described that there was no difference in transfusion rate in the trial.

Overall, 24 studies were included with 11 RCTs and 13 non-RCTs. For SSI, 18 studies with 8 RCTs and 10 non-RCTs were included in meta-analysis. The incidence of SSI in the follow-up periods was 1.2% (24/2067). We measured a composite outcome consisting of SSI, DVT, and revision TKA. We did not find any difference in the composite outcome between TKA-PSI and standard TKA groups: risk difference 0.00 (95%CI -0.01 – 0.01, p = 0.73), as shown in Fig. 11. In sensitivity analysis excluding non-RCTs, the result was consistent: risk difference 0.01 (95%CI -0.02 – 0.03, p = 0.46).

There was no significant difference in SSI between TKA-PSI and standard TKA groups: 13 out of 1049 patient in TKA-PSI versus 11 out of 1018 patients in standard TKA with risk difference 0.00 (95%CI -0.01 – 0.01, p = 0.83) as seen in Additional file 1: Appendix F. In sensitivity analysis excluding non-RCTs, the result was consistent: risk difference 0.01 (95%CI -0.02 – 0.03, p = 0.59) in RCTs.

For DVT, 16 studies were included in meta-analysis with 7 RCTs and 9 non-RCTs. The incidence of DVT was 1.5% (12/1716) in the pooled trials. There was no significant difference between groups: 9 out of 881 patients in TKA-PSI versus 3 out of 835 in standard TKA with risk difference 0.01 (95%CI -0.01 – 0.02, p = 0.28) as seen in Additional file 1: Appendix G. In sensitivity analysis excluding non-RCTs, the result was consistent: risk difference 0.01 (95%CI -0.01 – 0.03, p = 0.44) in RCTs. Pulmonary emboli in Boonen et al. (2016) and NCT 02539992 were counted as DVT, considering the overlap between pulmonary embolism and DVT.

For revision TKA, 14 studies were included and 13 studies were meta-analyzed with 4 RCTs and 9 non-RCTs. The incidence of revision TKA was 0.9% (11/1227) and there was no significant difference between groups: 3 out of 601 patients in TKA-PSI versus 8 out of 626 in standard TKA with risk difference − 0.01 (95%CI -0.02 – 0.01, p = 0.83) as seen in Additional file 1: Appendix H. The specific reasons for revision TKA were infection (n = 1), tibia loosening (n = 1), and patella resurfacing (n = 1) in TKA-PSI group, whereas infection (n = 2), tibia loosening (n = 2), patella resurfacing (n = 2), and instability (n = 2) in standard TKA group. Pourgiezis et al. (2016) experience one case with revision TKA in TKA-PSI group, but the data was not synthesized in meta-analysis, because the number of events in standard TKA was not shown. Also, Rathod et al. (2015) experienced reoperation for hematoma in TKA-PSI group, but we did not the case as revision TKA. Findings of the primary outcomes are summarized in ‘Summary of findings’ (Fig. 12).

Eight studies were included with the pooled 930 patients [45, 49, 53, 57, 58, 67, 68, 70]. Lower-limb alignment was monitored using hip-knee-angle (HKA) [53, 58, 67, 70] and mechanical axis [45, 49, 57, 68]. The positioning of the femoral prosthetic implant was examined using coronal [45, 49, 53, 57, 58, 67, 70] and sagittal [49, 53, 57, 58, 70] radiographs or CT scans. Also, rotational alignment of the femoral prosthetic implant was examined using CT scans [45, 58, 70]. The positioning of the tibial prosthetic implant was examined using coronal [45, 49, 53, 57, 58, 67, 70] and sagittal [45, 49, 53, 57, 58, 70] radiographs or CT scans. In the included studies, rotational alignment of the tibial prosthetic implant was not examined. All studies defined outliers as three or more degrees deviation from the planned alignment.

There were no significant differences in HKA (95%CI -0.09 – 0.05, p = 0.58), mechanical axis (95%CI -0.21 – 0.16, p = 0.81), femoral coronal positioning (95%CI -0.05 – 0.05, p = 0.95), femoral sagittal positioning (95%CI -0.13 – 0.14, p = 0.91), femoral rotational positioning (95%CI -0.32 – 0.07, p = 0.21), tibial coronal positioning (95%CI -0.03 – 0.05, p = 0.60), and tibial sagittal positioning (95%CI -0.16 – 0.10, p = 0.63) when comparing between TKA-PSI and standard TKA, as shown in Additional file 1: Appendix I.