The American Academy of Pediatrics (AAP) updated its guidance on male circumcision (MC) in 2012 to state “the procedure’s benefits justify access to this procedure for families who choose it.”¹ Whether MC should be considered an important public health intervention in the United States and other developed countries based on results of three randomized controlled trials^2–4 showing its HIV protective effect has been debated.^5–8 A key aspect of this debate is the rate of adverse events (AE), especially serious ones, attributable to MC, both for males circumcised as infants, and those undergoing voluntary circumcision as adults/adolescents.

Several studies have reported on MC AE, from mild to severe, ranging from 0.0008% to 3.6% in infants and from 0.9% to 8.8% in adults.^9–21 However, most of these studies²² were based on relatively small samples, one clinical site or state, cross-sectional data or non-representative cohorts. While the Weiss review ²² and one case series describing the experience of one pediatric urologist conducting Gomco circumcision in 150 neonates and infants²³ suggested generally higher rates of AE with older age at MC, to our knowledge, none have compared rates of AE across all age groups at MC from neonatal to adults in the same study. To provide stakeholders with better population-based information on the risk of MC AE, we use a large administrative claims dataset to 1) estimate the incidence rate of AE associated with MC via comparison of incidence risk ratio (IRR) and incidence rate difference (IRD) of AE between circumcised and uncircumcised newborn males, and 2) compare the IRR and IRD of AE associated with MC across age groups (≤1 year, 1 – 9 years, and ≥10 years).

SDIhealth (Plymouth Meeting, PA) consolidates U.S. electronic healthcare reimbursement claims. SDIhealth data include International Classification of Diseases, 9^th revision (ICD-9) and Current Procedural Terminology (CPT) codes, and are available about two months after clinical visits. SDIhealth creates a unique anonymous identifier for each patient, enabling individuals to be followed longitudinally.

The Charge Data Master (CDM) is SDIhealth’s inpatient dataset. It gathers data from a 20% convenience sample of all inpatient encounters of short-stay, acute care, and non-federal hospitals from 48 states and Washington DC, representing ~120 million unique hospitalized patients. CDM hospitals are located in all U.S. regions (25% East, 12% North, 45% South, 16% West, and data on regions is unspecified for 2%). Of these hospitals, 85% are urban, 36% are teaching, with a wide variability of bed-size (median size 200–299 beds). Of patients seen at these hospitals, about 10% of patients are covered by Medicaid, 30% by Medicare, and the remainder covered by Third Party payers. CDM data is formed by two datasets: CDM1 (~80% of CDM) is available since 2001 and updated monthly; only the month of diagnosis or procedure is provided, with date of discharge defaulted to the first day of the discharge month. CDM2 (~20% of CDM), is available since 2005 and updated weekly; unlike CDM1, the exact discharge date of a diagnosis or a procedure is available. SDIhealth also collects data from >870,000 unique outpatient medical providers with the exact discharge date of diagnoses and procedures. For this study, we used CDM data available through February 2010.

Possible MC AE for this analysis were first identified from a review of a) PubMed using the search terms “circumcision” and “adverse events,” and b) the ICD-9 and the CPT manuals for conditions that are not necessarily due to, but could be related to MC. Our search yielded 41 possible MC AE that we classified into ten clinical syndromic groups (eTable 1).

For each of the 41 possible MC AE, one of the co-authors who is a board-certified pediatric urologist (CSC), a priori defined the likely risk window in days based on pathogenesis (eTable 1). The possible AE were further classified by CSC as potentially serious (italicized in eTable 1) or not, based on clinical judgment, and assuming a worst case scenario.

We edited the CDM MC dataset by a) removing circumcised males who had a MC date prior to their birth date, and b) reclassified newborn males who did not have a MC record but had a MC-specific AE (CPT: 54162 and 54163) as circumcised.

We performed a retrospective cohort study using log binomial regression modeling (SAS 9.2) to ascertain the risk associated with MC. We first calculated the incidence of each AE over its risk window, per million circumcised (and separately for uncircumcised) newborn males using discharge date of circumcision (or birth for uncircumcised) for the beginning of the risk window. We then calculated the IRR, IRD, and their respective 95% confidence interval (CI) between the circumcised and uncircumcised groups. ²⁴

To minimize potential confusion on causal relationships in this exploratory study, the AE and person time outside the risk window in circumcised persons were deleted from analysis (instead of included in medical procedure-unexposed group as done in another risk window safety study).²⁵ An AE was considered probably related to MC if the IRR significantly exceeded one at p<0.05 or occurred only in circumcised newborn males. Multiple comparisons were not adjusted in our analysis because almost all the significant associations found were at p<0.001, which is less than any typically used correction factor (e.g., Bonferroni). To estimate the total incidence of AE associated with MC, we calculated the IRD between incidences of probable AE in circumcised vs. uncircumcised newborns, using unduplicated counts of males who had one or more AE in each group divided by the number of circumcised and uncircumcised newborns, respectively. For some syndromic groups, the risk window was not equal for all AE. To obtain the total for the syndromic group in this case, all conditions were followed for the longest risk window in the group. IRR and 95% CI were then generated.

We assessed if rates of probable MC AE differed in the following three age groups: males circumcised before one year of age (reference group), between one and nine years of age, and at age ten years or older. The age groups’ cut-off points separated infants from children prior to puberty and older males. We used the same statistical approach as above to calculate incidence per million male circumcisions (PMMC), IRR, IRD, and 95% CI.

To better detect rare MC AE, we first conducted the analysis using all available data including CDM1, CDM2, and outpatient datasets. Due to CDM1 day of discharge being defaulted to the first day of the month, all AE risk windows <28 days were reset to 28 days, the shortest risk window that could possibly be tracked, and the closest to a complete month, in this analysis (Tables 2 and 3). We then conducted a second analysis maximizing specificity of date by using only CDM2 and outpatient, the two datasets with exact dates for each procedure needed for exact risk window analysis (eResults, eTable2, eTable 3. This research was determined to be exempt from institutional review board evaluation because it entailed secondary analysis of administrative data procured from SDI Health (http://sdihealth.com/portal/site/imshealth) without personal identifiable human subjects. SDI Health originally collected this data from processing of US health care insurance reimbursement claims.

During 2001–2010, 1,400,920 MC reimbursement claims for males of all ages were submitted from U.S. hospital settings and available to SDIhealth (CDM1, CDM2, outpatient data). Forty seven males (0.0033%) had a MC dated prior to their birth date and these records were removed from analysis. Also, of all newborn males, 346 (0.015%) had a MC-specific AE but did not have a MC record. These were reclassified as circumcised newborn males.

We studied the AE outcomes after ~1.4 million MC in the United States, about 10 fold larger than the largest prior studies.^9–10 Using a broad definition of 41 possible MC AE to search a large medical administrative database, then restricting to the 16 probable MC AE with significantly elevated rates in pre-defined risk windows or occurring only in circumcised persons, we estimate the incidence of AE associated with newborn male circumcision in medical settings adjusted for the background rate to be less than half percent (0.30% for the more specific CDM2 dataset). Overall, the most common probable MC AE were related to correctional procedures at ~2000 PMMC and bleeding at ~1000 PMMC. Our findings were largely similar irrespective of whether the month-specific or date-specific datasets were used and consistent with the earlier U.S. studies given differences in methodology^{9, 10}.

Our findings also suggest that many AE such as penile reconstruction, pneumothorax and infections occur less frequently in circumcised males, perhaps due to a “healthy baby” bias – those newborns who undergo MC are more likely to be healthier (and without such disorders) compared to their uncircumcised counterparts. This type of selection bias is commonly seen in non-randomized observational studies of outcomes after medical procedures^{26, 27} and results in the observed lower rate of AE among circumcised males.

We found the incidence of MC AE was 10 – 20-fold higher when performed at older age groups compared to infancy. These findings are consistent with earlier studies^{22, 23} and may provide for the first time, a direct measure of the relative difference in AE rate by age at MC. Recent data on MC AE from a clinical trial in Kenya that included males 12 years of age or older, reported similar high rates of AE for this age group²⁸. Interestingly, in a study on infant MC AE from Kenya, an increased risk for AE was found if MC was performed in the second month of life compared to the first one²⁹. The indications for MC in older age groups in the U.S. may be more medical in nature (e.g., infections, adhesions) than the cultural/religious basis in most routine healthy newborns, however; future studies will need to carefully adjust for this potential source of confounding.

The incidence of amputations was highest among males circumcised at ten years of age or older: 0.17% (95% CI: 0.13 – 0.21%). In total, the absolute number of amputations in our database was 71. Most penile amputations captured in our dataset (45 out of 71) are recorded using ICD-9 code 643.0 which does not differentiate complete from partial penile amputation. Of the 71 recorded amputations, three were coded as complete — one among males circumcised in infancy, and two among males circumcised at ten years of age or older. Wiswell reported the absence of total penile amputations over five years in a study of MC AE among newborns from U.S. Army hospital settings⁹. Consistent with these findings, our data captured less than one total penile amputation PMMC, suggesting the possibility that most penile amputations recorded using ICD-9 code and captured in our dataset are likely to be partial. Without access to primary medical records, we can only speculate that the four patients that had penile amputation in the uncircumcised population likely were miscoded, were circumcised at non-medical settings, or patients undergoing operative intervention for severe genital anomalies. It’s noteworthy that other studies have reported on the success of treatment, including replantation, in the case of penile amputations.^{30, 31} We could not study deaths potentially related to MC as deaths in general are not captured in healthcare reimbursement claims databases like SDIhealth. In an earlier review, Wiswell reported three deaths due to male circumcision during the period 1954 – 1989 [~0.08 deaths from neonatal MC in the United States per year].⁹

Our study has several potential limitations. First, most of our data (~80%) assigns a discharge date of the first day of the month for the medical record. Hence, in the case where the AE has a risk window of < 28 days and falls in the same month of the MC it will be counted even if it occurs outside of the risk window. Also, in the case where an AE has a risk window of < 28 days and is encountered during the month following MC, it will be missed. The first scenario tends to over count some AE while the second tends to undercount some others. However, limiting our analysis to data with exact discharge date, our findings remained almost unchanged (eResults, eTable 2, eTable 3). At the same time, some of the males circumcised within the last year of our data, might have encountered an AE within a risk window outside of the available data. This might have decreased our overall rate of AE by a small fraction. Secondly, if an AE occurred on the same day of MC, it is impossible to determine whether the AE occurred before or after MC. Indeed, certain AE can also be an indication for MC. Hence, our reported rate might be inflated in case some AE were diagnosed on the same day as, or before MC.

A third limitation is that our data may not be generalized to the entire U.S. population as it came from a convenience sample. However,, the very large volume of administrative SDIhealth data used in this study (~20% of U.S. hospital discharges and >870,000 unique outpatient medical providers) strengthens our findings. A recent publication showed the trends in neonatal MC in SDIhealth data were virtually identical to that of two nationally representative datasets³² also further support its validity- at least for newborn males.

A fourth limitation of our data was that it was collected for billing purposes only. If a circumcision or an AE was not covered by a third party payer, it would be missing from this analysis. Also, some circumcisions might occur in non-medical settings, such as religious MC, but a resulting AE might require medical intervention, and hence be captured as occurring among uncircumcised newborns. Indeed, some uncircumcised newborn males in our data had a MC-specific AE. However, these did not exceed 0.01% of all newborns and the incidence of AE in our analysis was in the range of those from previous U. S. publications. ^9–10 Therefore, while the true rate may be lower or higher than our estimates, billing records should capture the vast majority of MC procedures.

Finally, MC can occur concurrently with other operative procedures for anesthesia-convenience reasons. The AE that might result from these cases might be confounded by other health conditions of the patient. Future studies overcoming these limitations and examining other databases to confirm our findings are needed to better estimate specific AE rates attributable only to MC.

Our data suggest the rate of AE associated with newborn circumcision is less than half percent. Importantly, the incidence of AE increased substantially when MC occurred after the first year of life. Given the current debate about whether MC should be delayed from infancy to adulthood for autonomy reasons³³, our results are timely and can help physicians counsel parents about circumcising their sons.