The PRISMA flow diagram for the systematic review is presented in Fig. 2. The initial search yielded 6878 potentially relevant articles. After removing 5014 duplicates and assessing titles/abstracts for eligibility, 1821 further articles were eventually excluded. Forty-eight studies were analyzed in full-text. Of these, 22 were excluded because the primary outcome of our review was not described. Finally, 26 articles were included in this systematic review and meta-analysis (Table 1) [2, 3, 7, 14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36].
Twenty-six included studies reported 5265 half-body examinations. Fourteen studies were performed during inguinal hernioplasty, 12 studies were performed during cadaveric dissections. A study by Mendes et al.  reported two different dissection types: Mendes2016a was for cadaveric dissection; Mendes 2016b was for during hernioplasty—which were counted as independent effect measures for the purposes of analysis. All studies were prospective in design.
Most studies were conducted in Europe (Table 2). Fifteen studies were performed in Europe, five in Asia, three in North America, two in South America (both studies from Brazil), and one in Africa. No studies were performed in Australasia (Table 3). The nerve identification rates at the inguinal canal were evaluated (Table 3).
Meta-analysis on the identification rate of the ilioinguinal nerve
Figure 3 shows the identification rate of the IIN. A total of 21 studies and 3773 half-bodies were analyzed using a random-effects model (Table 4). The overall identification rate was 94.4% (95% CI 89.5–97.9). In a leave-one-out sensitivity analysis, the identification rates varied slightly from 93.7 to 95.2%. The funnel plot and DOI plot indicated major asymmetry; therefore, we also examined fixed-effect model with heterogeneity correction.
Under the fixed-effect model with a heterogeneity correction, which gives more weight to large studies like Alfieri et al.  (n = 525) and Smeds et al.  (n = 973), the identification rate was 87.0% (95% CI 76.7%–95.8%) (Fig. 4). The median sample size for studies included in this analysis was 40. A follow-up unweighted multiple regression analysis showed that study sample size (β = − 0.74, p = .036) was the only statistically significant predictor for lower identification out of the following variables: sample size, year, region, number of centers, and type of dissection, (for the whole model: R2 = 0.56, F(8,12) = 1.94, p = .146). See Fig. 5 for a partial regression plot between sample size and PPE. The outlier in the bottom left of Fig. 5 was Lange et al. —a study with a small sample size (n = 40) and a low identification rate (75.0%, 95% CI 60.3%–87.4%). For the remainder of this analysis, we assumed that the sample size/prevalence relationship was a source of bias and, therefore, we described results for both random-effect and fixed-effect models.
Meta-analysis on the identification rate of the iliohypogastric nerve
Figures 6 (random-effects model) and 7 (fixed-effects model) show the identification rate for IHN. A total of 15 studies and 4187 half-bodies were analyzed. The overall identification rate for the IHN was 86.7% (95% CI 78.3%–93.3%) and 76.3% (95% CI 62.5%–88.9%) using a random-effects model and fixed-effects model, respectively. In a leave-one-out sensitivity analysis, the identification rates varied slightly from 84.4 to 88.1% for a random-effects model and from 74.3 to 80.9% for a fixed-effects model. A visual analysis of a funnel plot and DOI plot indicated marked asymmetry; we also noted that there was a negative relationship between prevalence and sample size—similar to the IIN outcome. Table 5 shows the results of the overall and subgroup analyses for the IHN. The identification rates ranged from 64.5% (95% CI 58.5%–70.4%) for multicenter studies to 99.9% (95% CI 99.1%–100.0%) for single center studies. As in the IIN outcome, there was a large, statistically significant amount of heterogeneity overall and within subgroups.
Meta-analysis on the identification rate of the genital branch of the genitofemoral nerve
Figures 8 and 9 and Table 6 show the results for the GNF. Fifteen studies and 3354 half-bodies were included. The identification rates for random-effects and fixed-effects models were 69.1% (95% CI 53.1%–83.0%) and 47.8% (95% CI 22.8%–73.0%), respectively. A leave-one-out sensitivity results ranged from 64.8 to 73.6% for a random-effects model and from 44.6 to 54.9% for a fixed-effects model. There was a large and statistically significant amount of heterogeneity overall and within subgroups. As with other outcomes, a visual analysis of funnel and DOI plots indicated irregularity and provided evidence that larger studies tended to have smaller prevalence.
Meta-analysis of anatomical reference points
Table 7 presents all secondary endpoints including the pooled estimates of distance of the point of the nerve emergence in relationship to the anatomic landmarks for the IIN, which was located inferior to the ASIS, medially to the ASIS, and the inguinal ligament. In one study , the range was reported instead of the standard deviation. Therefore, we estimated the standard deviation from the range using the guidelines in Hozo et al.  assuming an underlying normal distribution of nerve lengths. The distance from IIN emergence inferior to the ASIS was 2.8 cm (2.65–2.95) and medially to the ASIS was 3.62 cm (3.04–4.19). For nerve length outcomes, there was a large and statistically significant amount of heterogeneity. When computable, the heterogeneity estimates for other reference points were also large and statistically significant.