After screening a total of 1412 articles, 93 original works published in English and reporting on a minimum of 10 patients treated with stereotactic or robotic thermal ablation for malignant liver tumours were included for full text screening. After additional exclusion of 59 works, 34 articles were included for qualitative analysis and 22 for quantitative analysis (Figure 1).
All of the 34 included works were single-centre studies, of which 26 were retrospective studies(29,30,39–48,31,49–54,32–38), 3 prospective case series(55–57), 3 prospective cohort studies(58–60) and 2 randomised controlled trials(61,62). Two studies reported results using a laparoscopic treatment access(44,55), the other 32 reporting on thermal ablations using a percutaneous approach. Six studies applied robotic targeting using mechanical tracking(31,47,48,51,60,62), 7 an electromagnetic (EM)-tracked dynamic technique(35,42,50,55,57,59,61) and the other 21 an optically tracked stereotactic aiming device. Baseline study, patient and lesion characteristics, applied ablation and navigation techniques and reported outcomes of the included works are shown in Table 2. The published literature on stereotactic or robotic guidance for thermal ablation of liver tumours increased continuously since the first clinical series in 2011. This was the case regarding all reported endpoints, and most prominently for safety and treatment efficacy, as illustrated in Figure 2.
Targeting accuracy
Of the 34 included studies, 10 reported on targeting accuracy, of which nine reported Euclidean, lateral or angular targeting errors after stereotactic ablation probe positioning. The remaining study reported a 95.6% targeting accuracy, defined as the centre of the ablated zone being located within a 5 millimetre range from the preoperatively defined ideal target point, as assessed on a 24h CT/MRI scan(50). Pooled estimates for Euclidean, lateral and angular targeting errors were 5.3 (95% CI 3.9, 6.7), 3.7 (3.0, 4.4) and 2.4 (1.7, 3.1) millimetres, respectively, with significant between-study heterogeneity (Table 3). Summary estimates for lateral targeting errors are displayed in Figure 3.
Three studies investigated factors influencing targeting accuracy, of which two studies performed univariable between-group analyses and one group used multivariable linear regression. Mauri et al.(50) showed no influence on “correct targeting” (defined as the "Centre of the ablated zone located within 5-mm range from the ideal target point preoperatively established", as assessed on CT/MR imaging 24 hours after ablation) by tumour entity, lesion characteristics and applied guidance and ablation techniques. Widmann et al.(56) described larger lateral errors at the ablation probe tip in lesions located in “subphrenic plus fat” as opposed to “clear parenchymal” positions and in Segment II versus Segment IV. Tinguely et al.(29) reported statistically significant higher lateral targeting errors with raising targeting trajectory lengths (0.2 millimetre per additional centimetre) and when targeting tumours in cirrhotic livers (by 0.7 millimetre) in a multivariable model, with no influence shown for challenging intrahepatic lesion locations or more complex targeting trajectories.
Procedural efficiency and safety
Eighteen studies reported on procedural efficiency and all included works reported on safety related to stereotactic targeting for thermal ablation. The need for readjustment of ablation probes due to insufficient accuracy were reported varyingly across studies. Four studies reported mean numbers of ablation probe readjustments of n = 0(62), 0.8(48), 1.1(60) and 2.4(61). Nine authors reported relative numbers of probe readjustments per patients (4.8%(35), 5.6%(58), 35%(51) and 60%(48)) or per lesions (1%(29), 4%(45), 5.6%(58), 8.8%(55) and 41.2%(47)). Mean overall procedure duration ranged between 18.3 and 254.5 minutes in 12 studies, and navigated targeting duration between 1.5 and 36.3 minutes in 8 reporting studies. Mean total DLP ranged between 807 and 2216 mGy*cm in 11 studies reporting overall radiation exposure. Hospital length of stay ranged between 0 and 7 day in 18 reporting studies.
All included studies reported on treatment related complications, applying varying types of definitions. Twelve studies used the definitions proposed by the Society of Interventional Radiology (SIR)(63), 8 applied the Clavien-Dindo classification(64), 2 the definitions proposed by Ahmed et al(24), one the CIRSE classification and 11 studies used other definitions or did not further specify. The overall complication rate ranged between 0 and 57.9%, the pooled estimate being 11.4% (CI 6.7, 16.1; I2 87.9%, p < 0.01) in 16 studies included for meta-analysis. Major complications ranged between 0 and 20.5%, the 20.5% rate being reported in a study of patients undergoing stereotactic radiofrequency ablation for very large (≥ 8 cm) tumours(33). The overall pooled estimate for major complication rate was 2.4% (CI 1.4, 3.6; I2 20.7%, p = 0.198), which was lower in the subgroup applying the Clavien-Dindo classification (2.0%, CI 0.7, 4.0) versus the subgroup applying the SIR classification (4.0%, CI 1.0, 8.8) (Figure 4). Mortality rates ranged between 0 and 4.3%, with a pooled estimate of 0.8% (CI 0.4, 1.4; I2 0%, p 0.99) in 20 studies included for meta-analysis.
Treatment efficacy
Thirty-one of the 34 included studies reported on treatment efficacy. Varying definitions were reported for all treatment efficacy outcomes, including the time points for assessment and the duration of follow-up. Six studies(29,31,41,42,53,61) referred to the terminology for follow-up assessment after ablation of liver tumours proposed by Ahmed et al(24), while thirteen studies applied similar definitions without explicitly referencing this classification. Individual descriptions and time points of follow-up assessments are available for each study in the Supplementary File. The varying time points and durations of follow-up assessments in studies reporting treatment efficacy are illustrated in Figure 5.
Keeping in mind this variability in definitions as well as the varying specific inclusion criteria for patients and lesions in several studies (e.g. very large tumours ≥ 8cm(33), vanishing lesions(30)), reported rates for technical success ranged from 90.2 to 100% (18 studies), for primary technique efficacy from 80.5 to 100% (27 studies) and for secondary technique efficacy from 90.2 to 100% efficacy (13 studies) and for LTP from 0 to 54% (21 studies). Quantitative analysis of primary technique efficacy (i.e. complete tumour ablation at the first follow-up imaging) according to time points of the first follow-up is summarised in Figure 3. Primary technique efficacy rates were reported higher in studies performing a first follow-up after 1 to 6 weeks (pooled estimate 93.6%, CI 88.9, 97.1), than in studies assessing primary technique efficacy at 6 to 12 weeks (pooled estimate 90.1% (CI 87.2, 92.7). Despite subgroup analysis, a statistically significant between-study heterogeneity remained in the former group (Figure 6).
Six studies presented analyses on factors influencing treatment efficacy when using stereotactic tumour targeting, four of which included multivariable regression analyses. Tinguely et al.(29) showed a statistically significant influence of tumour size (</> 3cm) and targeting accuracy (</> 5mm) on LTP, with no influence by more complex intrahepatic tumour locations, in a multivariable model accounting for clustering by using Generalized Estimating Equation (GEE). Schaible et al. reported tumour size (</> 3cm) next to the type of targeting approach (stereotactic vs. free-hand) to be a significant predictor of primary technique efficacy in a similar logistic GEE model(31). Lachenmayer et al.(41) found vessel proximity and tumour size (</> 3cm) as independent predictors for LTP. Hirooka et al.(59) reported the free-hand as opposed to the stereotactic approach to be significantly associated with “local residual recurrence” in Cox regression analysis. In univariable between-group comparisons, Widmann et al.(54) reported differences in “technique effectiveness” for lesion size </> 5cm and hollow viscera vicinity and no differences for tumour entity and lesion location. Bale et al.(52) found differences in “local recurrence” rates for lesions in proximity to vessels, bile ducts and hollow organs.
Comparative studies
Nine studies compared stereotactic versus “free-hand” ablation for varying endpoints, including two randomised controlled studies(61,62) and three prospective cohort studies(59,60) of which one matched pair-analysis(58). Main study characteristics and reported results for targeting accuracy, procedural efficiency and safety and treatment efficacy are summarised in Table 4.
Targeting accuracy was shown to be significantly enhanced when using stereotactic targeting in 3 out of 4 studies (one of them after manual adaption of the ablation probe(47)). The randomised controlled trial by Heerink at al.(62) confirmed enhanced accuracy specifically for out-of-plane trajectories (5.9 versus 10.1mm), and showed a significant reduction of ablation probe repositionings in robotic versus free-hand ablations (0 versus 1, primary study endpoint). This was confirmed by Zhang et al.(61) showing fewer instrument readjustments (2.4 versus 4.95) when using EM-guided targeting, and by Mbaliske et al.(60) when using robotic as opposed to conventional CT guidance (1.1 versus 3 readjustments).
Durations for overall procedures and for ablation probe positionings were reported variably across studies. Zhang et al.(61) showed a significant reduction in number of CT scans used for interventions (7 vs. 10), in CT fluoroscopy time and in total DLP when using EM guided as opposed to free-hand targeting. Four other studies showed a reduction, and 2 studies showed an increase in total DLP in the stereotactic versus free-hand cohort using (Table 2). In the RCT of Heerink et al.(62), the number of navigational CT scans tended to be lower in stereotactic versus free-hand procedures (5 versus 7), and the increase in DLP was presumably due to the larger scan field necessary to include the optical reference fiducials.
With respect to treatment efficacy, Zhang et al.(42) showed a higher complete ablation rate in the first session (3 days after ablation) when using stereotactic EM guidance, but equal primary technique efficacy at 1 month. Schaible et al.(31) reported a significantly higher primary technique efficacy when using robotic targeting in a retrospective comparative study. The pooled odds ratio for stereotactic versus free-hand targeting was 1.94 (CI 1.18, 3.19) (Figure 7). All six studies included in this meta-analysis assessed primary technique efficacy at a first follow-up imaging between 1 and 6 weeks.
Four propensity score matched analyses compared stereotactic thermal ablation for different subtypes of patients or lesions. Stereotactic RFA yielded similar safety and treatment efficacy outcomes when comparing CT ‘invisible’ tumours ablated with MR image fusion to CT visible tumours(30), octogenarians to a younger study population(36), lesions in a subphrenic location to non-dome locations(39) and HCC in a subcardiac position to a non-subcardiac location(40).