Conflict resolution of the beams: CT vs. MRI in recurrent hernia detection: a systematic review and meta-analysis of mesh visualization and other outcomes

We performed a systematic search of online databases, including PubMed, Scopus, Embase, and Web of Science, from inception up to July 12, 2024. The search terms were designed to capture key concepts related to mesh visualization for detecting recurrent hernias using MRI or CT. Our study commenced on the third of July 2024. The used search strategy was structured as follows:

(Computed Tomography OR CT OR Computed Tomography Angiography OR IVCT OR low-dose CT OR Magnetic Resonance OR Magnetic Resonance Imaging OR MRI OR Magnetic Resonance Angiography OR contrast MRI OR IVMRI OR Diagnosis OR Diagnose) AND (Recurrent hernia OR hernia recurrence OR recurrence of Hernia). Figure 1 presents a word cloud visually representing the search terms and their relevance, emphasizing the primary focus areas of our systematic review.

Bild vergrößern

This study adheres to the Meta-Analysis of Observational Studies in Epidemiology (MOOSE) Checklist [23], and a supplementary table (Supplementary Table 1) is provided to outline compliance with each item. Our meta-analysis adhered to the PICOS (Participants, Intervention, Comparator, Outcomes, and Study Design) framework to ensure methodological rigor. The participant population included patients undergoing imaging for recurrent hernia evaluation, as defined by the inclusion criteria. The interventions assessed were CT and MRI modalities, with direct comparisons between the two imaging techniques. The baseline characteristics of the included studies are presented in Table 1.

Key outcomes included recurrence rates, mesh visualization, seroma detection, and reoperation rates. In this study, the MINORS instrument was applied to evaluate the methodological quality of our studies. This assessment included criteria such as clearly stated aims, consecutive patient inclusion, unbiased endpoint assessment, and adequate follow-up, ensuring a standardized quality evaluation for all studies as per Table 2.

Additionally, we ensured alignment with key components of the AMSTAR guidelines, including a comprehensive search, independent review, and quality assessment of included studies. Furthermore, MeSH and Emtree terms were utilized in the database search, and these terms are detailed in Supplementary Table 2.

We included studies evaluating the recurrence of external hernias in adults, using MRI or CT to visualize biological or synthetic meshes. Eligible studies included full texts, abstracts, and letters published in peer-reviewed journals, limited to the English language. Excluded were studies using ultrasonography or alternative imaging, addressing internal hernias or unrelated conditions, as well as animal studies, case reports, Meta-analyses, and reviews. Studies focusing on internal hernias, including parastomal hernias, were excluded to ensure the scope of the analysis remained on external hernias. The types of included studies are listed in Supplementary Table 3.

Four authors independently conducted the screening process, starting with titles and abstracts, followed by a full-text data extraction. Studies were assessed according to the predefined inclusion and exclusion criteria. Ultimately, 26 studies were selected for inclusion, with 8 studies in the MRI group [5, 24‐30] and 18 studies in the CT group [31‐48]. Data extraction was then performed to collect baseline characteristics and outcome measures from the included studies, facilitating further analysis.

Two reviewers assessed bias in the included studies using the MINORS criteria [49], which rates non-comparative studies on a 0–16 scale and comparative studies on a 0–24 scale. Higher scores reflect lower bias risk. Non-comparative studies were categorized as extremely poor (0–4), low (5–7), fair (8–12), or excellent quality (13–16). Similarly, comparative studies were rated as very poor (0–6), bad (7–10), good (11–15), or excellent quality (16–24).

The GRADE (Grading of Recommendations Assessment, Development, and Evaluation) system was used. The GRADE system was applied to each outcome included in our meta-analysis [50], considering factors such as study design, result consistency, estimation precision, potential biases, and clinical relevance of the findings.

A meta-analysis of proportions was conducted to assess recurrence rates, seroma detection, mesh visualization, and reoperation needs in hernia repair, comparing CT and MRI detection methods. The analysis utilized R Studio (version 2024.09.0, Build 375) with relevant statistical packages [51, 52]. A random-effects model was applied to address heterogeneity among studies [53]. A random-effects model using the DerSimonian-Laird method was applied to account for between-study heterogeneity. Given the substantial heterogeneity observed (I² = 89%, τ² = 0.0085, p < 0.01), the between-study variance component (τ²) influenced the weighting calculation, leading to a more uniform distribution of study weights compared to a fixed-effect model. [54‐56]. While the study’s weight is substantial in the fixed-effect model (99.2%), the random-effects model appropriately adjusts for the high between-study heterogeneity, resulting in a more balanced weight distribution across studies [54]

Effect sizes were calculated as untransformed proportions, representing event rates in each study. A continuity correction of 0.5 was applied to studies with zero events to ensure valid calculations of proportions and confidence intervals. Heterogeneity was evaluated using the I² statistic, quantifying variability among studies [55].

The use of untransformed proportions with the DerSimonian-Laird estimator effectively addresses between-study heterogeneity while preserving the direct clinical relevance of our findings.

Subgroup analysis compared CT and MRI using the Chi-squared (X2) test [56], and the DerSimonian-Laird method was employed to estimate between-study variance (τ2), with studies weighted by the inverse variance method [57, 58]. Results were displayed in forest plots, showing study proportions, subgroup pooled proportions, and overall proportions with 95% confidence intervals [59]. A p-value < 0.05 was deemed statistically significant for all tests [60]. We conducted Egger’s regression test for funnel plot asymmetry across all outcomes. Egger’s test evaluates the relationship between study size and effect size precision, where a statistically significant p-value (p < 0.05) indicates potential small-study effects or publication bias.

A comprehensive database search initially identified 3,124 records. After removing duplicates, 2,466 references remained for title and abstract screening. Following a thorough screening process and full-text review, 26 studies involving a total of 3,725 patients were included in this systematic review and meta-analysis. The PRISMA flow chart of the selection process is shown in Fig. 2 [61].

Bild vergrößern

Table 1 highlights the baseline characteristics of the included studies and the differences between MRI and CT-based studies in recurrent hernia evaluation. Eight MRI studies, comprising 245 patients, were included, compared to 18 CT studies with a total of 3,480 patients. Patient demographics showed differences in gender distribution, with MRI studies reporting 140 males and 55 females (71.8% male) among studies with gender data. CT studies reported 697 males and 440 females (61.3% male) where gender data was available. The mean patient age was comparable between modalities, with MRI patients averaging 58.7 ± 12.2 years and CT patients averaging 59.3 ± 11.2 years.

In this study, outcomes were categorized into primary and secondary outcomes to structure the analysis. The primary outcomes included recurrence rates and mesh visualization, as they directly addressed the study’s main objective. The secondary outcomes comprised seroma detection and reoperation rates, providing additional insights into postoperative complications.

The analysis of recurrence rates using a random-effects model showed an overall estimated proportion of 20% (95% CI: 0% to 42%) for CT and 15% (95% CI: 4% to 26%) for MRI, with high heterogeneity in both (I² = 100%, τ² = 0.1937, p < 0.01 for CT; I² = 82%, τ² = 0.0143, p < 0.01 for MRI). The subgroup analysis, based on 15 studies for CT and 6 studies for MRI, showed no significant variation in recurrence rates between CT and MRI (χ²₁ = 0.13, p = 0.72) (Fig. 3a).

Bild vergrößern Bild vergrößern

Sensitivity analysis highlighted the variability in CT recurrence rates, ranging from 14% (95% CI: 9% to 18%) when Liang et al. 2012 [36] was excluded to 21% (95% CI: −2% to 44%) when Kumar et al. 2022 [37] was removed (Fig. 3b). For MRI, rates ranged from 9% (95% CI: 2% to 16%) when Paajanen et al. 2004 [28] was excluded to 19% (95% CI: 4% to 34%) when Musters et al. 2016 [27] was removed (Fig. 3c).

The overall analysis showed a mesh visualization rate of 48% (95% CI: 0% to 100%) for CT and 73% (95% CI: 42% to 100%) for MRI, both with high heterogeneity (I² = 100%, τ² = 0.3139, p < 0.01 for CT; I² = 95%, τ² = 0.0911, p < 0.01 for MRI). Based on 4 studies each for CT and MRI, the subgroup analysis showed no significant variation in mesh visualization rates between the two modalities (χ²₁ = 0.58, p = 0.44) (Fig. 4a).

Bild vergrößern

Analysis of CT mesh visualization rates revealed a range from 31% (95% CI: 6% to 55%) when Liang et al. (2012) [36] was excluded, to 61% (95% CI: 14% to 107%) when Kumar et al. (2020) [37] was excluded (Fig. 4b). For MRI, the rates ranged from 63% (95% CI: 24% to 103%) when either Köhler et al. (2015) [25] or Musters et al. (2016) [27] was excluded, to 85% (95% CI: 57% to 113%) when Fischer et al. (2007) [24] was excluded (Fig. 4c).

The analysis of seroma detection rates highlighted variations between CT and MRI modalities. The overall estimated seroma detection rate for CT was 12% (95% CI: 4% to 19%), based on 8 studies, with high heterogeneity (I² = 90%, τ² = 0.0088, p < 0.01). In contrast, MRI showed an overall estimated detection rate of 10% (95% CI: 4% to 15%), based on 3 studies, with low heterogeneity (I² = 11%, τ² = 0.0004, p = 0.33). The subgroup analysis revealed no significant differences in seroma detection rates between CT and MRI (χ²₁ = 0.20, p = 0.65) (Fig. 5a).

Bild vergrößern

In the sensitivity analysis for CT seroma detection, the highest rate was observed when Liang et al. 2012 [36] was removed (14% [95% CI: 7% to 20%]), while the lowest rate was noted when Sasse et al. 2018 [41] was excluded (11% [95% CI: 3% to 18%]) (Fig. 5b). For MRI seroma detection, the highest rate in the sensitivity analysis was observed when Paajanen et al. 2004 [28] was removed (15% [95% CI: −3% to 33%]), and the lowest when Van den Berg et al. 2000 [30] was excluded (9% [95% CI: 4% to 14%]) (Fig. 5c).

The analysis of reoperation rates revealed variations between CT- and MRI-based studies. The overall estimated reoperation rate for CT was 6% (95% CI: 1% to 11%), based on 3 studies, with moderate heterogeneity (I² = 72%, τ² = 0.0013, p = 0.03). In comparison, MRI had a higher overall estimated reoperation rate of 34% (95% CI: 3% to 66%), also based on 3 studies, with high heterogeneity (I² = 93%, τ² = 0.0688, p < 0.01). Although the test for subgroup differences suggested a trend towards variation in reoperation rates between CT and MRI, it did not reach statistical significance (χ²₁ = 3.04, p = 0.08) (Fig. 6a).

Bild vergrößern

During the sensitivity analysis for CT reoperation rates, the highest rate was observed when Holihan et al. (2016) [34] was excluded, resulting in 8% (95% CI: 1% to 15%), while the lowest rate occurred when Sasse et al. (2018) [41] was excluded, yielding 3% (95% CI: 1% to 5%) (Fig. 6b). For MRI reoperation rates, the highest value was recorded when Kirchhoff et al. (2010) [26] was excluded, at 51% (95% CI: 38% to 64%), and the lowest was noted when Paajanen et al. (2004) [28] was excluded, at 22% (95% CI: −4% to 48%) (Fig. 6c).

The overall certainty of evidence was rated high for recurrence and reoperation rates in both MRI and CT studies, as Egger’s test indicated no significant publication bias (p > 0.05), and the included studies directly addressed the research question. However, for mesh visualization, the certainty of evidence was downgraded to low due to significant publication bias observed in both MRI (p = 0.0082) and CT studies (p = 0.0028), indicating potential small-study effects.

Similarly, the certainty of evidence for CT seroma detection was rated low due to a significant Egger’s test result (p < 0.0001). In contrast, MRI seroma detection rates were supported by high-certainty evidence, as no publication bias was detected (p = 0.7671). Despite the overall precision of the effect estimates, high heterogeneity (I² > 90% in some outcomes) led to further downgrading of evidence for mesh visualization and reoperation rates as shown in supplementary Tables 5, 6.

Significant heterogeneity was noted across CT- and MRI-based studies for hernia evaluation outcomes. For mesh visualization rates, heterogeneity was substantial (I² = 99%, τ² = 0.2324, p < 0.01) (Supplementary eFigure 1a). Subgroup analysis by contrast usage showed persistently high heterogeneity across all subgroups (I² > 88%) (Supplementary eFigure 1a–d), indicating that contrast agent variations alone do not explain the variability. Subgrouping by hernia type revealed differences in heterogeneity levels (Supplementary eFigure 2a–d).

Leave-one-out sensitivity analysis reduced heterogeneity significantly for CT reoperation rates, eliminating it (I² from 72 to 0%) when Holihan et al. (2016) [34] was excluded (Supplementary eFigure 3a). For CT seroma detection, excluding Schoenmaeckers (2010) [40] reduced heterogeneity from 90 to 70% (Supplementary eFigure 3b). In MRI studies, recurrence rate heterogeneity decreased (I² from 82 to 57%) when Musters et al. (2016) [27] was excluded (Supplementary eFigure 4a), and for MRI reoperation rates, excluding Paajanen et al. (2004) [28] reduced heterogeneity to 0% (I² from 93 to 0%) (Supplementary eFigure 4b).

These analyses identified Musters [27], Paajanen [28], and Holihan [34] as key contributors to heterogeneity (Supplementary eFigures 3a, 4a, and b). In contrast, sensitivity analyses for mesh visualization rates and CT recurrence rates showed minimal to no impact on heterogeneity (Supplementary eFigures 3c, d, and 4c).

Regarding publication bias, funnel plots were generated for mesh visualization, need for reoperation, recurrence rates (Supplementary eFigure 5a–c), and seroma detection (Supplementary eFigure 5d). On visual inspection, asymmetric distribution was noted for mesh visualization and recurrence rates (Supplementary eFigure 5a and c), with several studies falling outside the expected funnel boundaries, indicating possible publication bias. For the need for reoperation and seroma detection rates (Supplementary eFigure 5b and d), although some asymmetry was observed, most studies fell within the expected funnel boundaries, suggesting less likelihood of publication bias for these outcomes. In order to assess the publication bias using Egger’s test, the p-values for each outcome were as follows: recurrence in MRI studies (p = 0.4203) and CT studies (p = 0.4777) indicated no significant publication bias. For mesh visualization, significant evidence of publication bias was found in both MRI studies (p = 0.0082) and CT studies (p = 0.0028). The seroma outcomes showed no significant publication bias for MRI studies (p = 0.7671), whereas CT studies showed strong evidence of publication bias (p < 0.0001). Lastly, for the need for reoperation, both MRI (p = 0.8837) and CT (p = 0.6865) studies demonstrated no significant publication bias as per Supplementary Table 5, 6.

Heterogeneity posed a significant challenge in analyzing CT-based outcomes due to variability in study designs, populations, and methodologies. CT recurrence rates showed high overall heterogeneity (100%), with a modest reduction to 98% in non-contrast studies. Subgrouping by hernia type further reduced heterogeneity, notably to 77% for incisional hernias and 93% for ventral hernias, though it remained high for inguinal (100%) and other hernia types (96%).

Similarly, for CT mesh visualization, heterogeneity was 100% overall but decreased to 94% with intravenous contrast. Subgrouping by hernia type led to significant reductions, particularly for inguinal hernias (0%), ventral hernias (91%), and incisional hernias (96%). For CT seroma detection, overall heterogeneity was 90% but dropped to 70–78% depending on hernia type. Unexpectedly, subgrouping non-contrast studies for CT reoperation rates increased heterogeneity to 90%, while grouping by hernia type reduced it to 67–98%. These findings highlight the importance of standardized imaging protocols and consistent reporting to reduce variability in future research.

Given the significant heterogeneity in our analysis, we believe the random-effects model provides a more appropriate representation of the true uncertainty in our estimates [71].