Results of search strategy
A total of 3,366 papers were identified from the database searches and the full text was obtained for 306 of these (Fig. 1). A total of 269 papers were excluded based on the full text, including two studies that compared the prevalence of circumcision among cases of invasive penile carcinoma with cases of in situ penile cancer. An additional six papers were identified from references and review papers. Eight eligible papers were identified which included information on penile cancer.
Description of eligible papers
The eight papers evaluating the effect of circumcision on penile cancer included one cross-sectional study and seven case–control studies (Table 1). Study populations were from Europe (n = 4), the United States (n = 3), and China (n = 1).
Age at circumcision was categorised into three groups: childhood/adolescence (all participants < 18 years); mostly adults (median age ≥ 18 years; minimum 10 years old); and age inadequately reported (Table 1). The proportion of men circumcised during childhood/adolescence ranged from 4% in Denmark  to 61% in the United States . Circumcision among adults ranged from 0.7% in China  to 5% in the USA .
Five papers evaluated the effect of circumcision on PIN and/or in situ cancer [23, 24, 28, 30, 31] and six papers evaluated the effect on invasive cancer [14, 23, 24, 27, 28, 32] (Fig. 1; Table 1).
Quality of studies
One study had five quality domains classified as low risk and two domains classified as unclear risk ; all other studies had a high risk of bias in one or more of the quality domains (Table 2).
The potential for length bias was considered to be high for three case–control studies, where prevalent cases were identified and ≥20% could not be contacted for an interview [23, 24, 28] and unclear in one other study  (Table 2). For these studies, the proportion that were known to have died between diagnosis and the time of the study was 13% , 17%  and 34–38% .
The risk of bias from poor participation was low for only one study  (Table 2). Fewer than 80% of identified subjects agreed to participate for three studies [23, 24, 28] and for one of these, the proportion that agreed to participate was substantially higher for controls (60%) compared to cases (27%) . Information on participation was not available in three studies [30–32].
In four studies, the age at circumcision was not reported or incompletely reported, [24, 30–32] and the temporality of any observed association between circumcision and penile cancer could not be assessed. In a fifth study , although the exact age at circumcision was not reported, the authors report that none of the boys were circumcised at birth and all participants were circumcised more than 5 years prior to the study. A high risk of confounding was present in three studies that did not adjust for age in either the study design or analysis [14, 30, 31].
Association between circumcision and penile cancer
Penile intra-epithelial neoplasia and in situ penile Cancer
Seven analyses evaluated the association of circumcision on PIN and in situ penile cancer [23, 24, 28, 30, 31] (Table 1; Fig. 1).
One of the three analyses of circumcision in childhood/adolescence and PIN or in situ penile cancer compared men circumcised in childhood/adolescence with men not circumcised in childhood/adolescence  and was thus excluded from the meta-analysis. In this study, there were no penile cancer cases among men circumcised in childhood/adolescence, but the 95% CI was wide (OR = 0.00, 95% CI 0.00–5.96). The remaining two analyses were deemed at low risk of bias from confounding in our quality assessment. These showed little evidence of an association of circumcision with PIN and in situ penile cancer (summary OR = 0.93, 95% CI 0.59–1.47; p-heterogeneity = 0.86; I
2 = 0.0%) (Fig. 2). The OR from the single analysis among men circumcised mostly as adults was smaller (OR = 0.53; 95% CI 0.06–2.28) but confidence intervals were wide, and there was little evidence of a difference in this effect estimate and the summary estimate for those circumcised in childhood/adolescence (p = 0.60).
Three analyses did not adequately report age at circumcision. For one (a cross-sectional study from France), circumcision was likely to have occurred in childhood and this showed some evidence of a protective effect of circumcision on PIN and in situ penile cancer (OR = 0.56; 95% CI 0.30–1.00) . A sensitivity analysis including this analysis with the other analyses of childhood/adolescent circumcision provided only weak evidence of a protective effect on PIN/in situ penile cancer (summary OR = 0.77; 95% CI 0.54–1.11 p-heterogeneity = 0.42; I
2 = 0.0%; n = 3).
Invasive penile cancer
Nine analyses from six papers evaluated the effect of circumcision on invasive penile cancer [14, 23, 24, 27, 28, 32]; (Table 1; Fig. 1). Three of the four analyses of circumcision in childhood/adolescence, used a comparison group of never circumcised men. The overall OR from these three analyses showed strong evidence of a protective effect (summary OR = 0.33; 95% CI 0.13–0.83) (Fig. 3). There was some evidence of heterogeneity in estimates (p-heterogeneity = 0.11; I
2 = 55%) although all three studies showed a protective rather than detrimental effect. The sensitivity analysis including only the two analyses that adjusted for covariates showed little evidence of heterogeneity (summary OR = 0.50; 95% CI 0.29–0.86; p-heterogeneity = 0.69; I
2 = 0.0%; n = 2).
The relationship was reversed for analyses of men circumcised mostly as adults (summary OR = 2.71; 95% CI 0.93–7.94; p-heterogeneity = 0.15; I
2 = 48%; 3 analyses).
Two analyses inadequately reported the age at circumcision and neither provided evidence of a protective effect of circumcision on invasive cancer (Table 1; Fig. 3). In one of these analyses, from a case–control study from the UK, the one cancer patient who was circumcised had been circumcised in childhood, and the controls are also likely to have been circumcised during childhood . Including this analysis in the summary estimate for childhood/adolescent circumcisions gave similar results (summary OR = 0.36; 95% CI 0.17–0.75; n = 4) to the overall summary at these ages (OR = 0.33; 95% CI 0.13–0.83; n = 3) with less heterogeneity in estimates (p-heterogeneity = 0.22; I
2 = 32% vs. p-heterogeneity = 0.11; I
2 = 55%).
For the two analyses [23, 24] which assessed the effect of childhood/adolescent circumcision on both in situ and invasive penile cancer, there was a stronger association with invasive disease (OR 0.53 vs. 0.91 in the Daling et al.  paper and 0.41 vs. 1.0 in the Tseng et al.  paper) (Table 1). However, a meta-regression of the five analyses evaluating the effect of childhood/adolescent circumcision compared to no circumcision revealed weak evidence that the effect differed for in situ and invasive penile cancer (p = 0.20), although power for this analysis was low [14, 23, 24].
The role of phimosis in the association between circumcision and penile cancer
To examine the association of circumcision with penile cancer beyond the effect mediated through phimosis, two papers stratified analyses by history of phimosis [23, 24]. Both of these found an association of circumcision with invasive penile cancer among all subjects, but no evidence of an association among those with no history of phimosis (Table 3).