THE CIRCUMCISION REFERENCE LIBRARY
Presented in part at the Annual Meeting of The Society for Pediatric Research in Anaheim, California, on May 10,1990.
Address correspondence to: John B. Chessare, M.D., M.Ph., Department of Pediatrics, Medical College of Ohio, P.O. Box 10006, Toledo, Ohio 43699
Recent information regarding the increased risk of urinary tract infections in the first year of life for uncircumcised boys has created confusion regarding the appropriate guidance to be given to parents confronting the circumcision issue. A decision model was built that addressed the question of whether or not to circumcise a newborn male considering the probability of a non-circumcised boy having a UTI in the first year of life (0.041), the probability of a circumcised boy having a UTI in the first year of life (0.002), and the likelihood of renal scarring from a UTI (0.075). After considering the morbidity associated with the procedure, all possible outcomes were ranked from worst to best (circumcised-renal pathology to uncircumcised-no infection) and given a value on a 0 to 1 scale. For the set of values assigned to the outcomes, the choice of no circumcision yielded the highest expected utility. For the set of assigned utilities, sensitivity analysis showed that unless the probability of a UTI in the first year of life for an uncircumcised male was greater than or equal to 0.29, then non-circumcision was still the preferred choice. The decision was most sensitive to the degree of aversion to the morbidity associated with the procedure (pain, bleeding, inflammation).
Circumcision is performed on 60% to 90% of newborn males in the United States.1,2 Wallerstein has estimated that 20% of circumcision in this country is carried out for religious beliefs. The remainder is done for socio-cultural and health reasons. It has also been pointed out that 80% of the world's population does not practice circumcision, with Japan, Germany, and the Scandinavian countries among the medically advanced societies opting against this procedure.1
The medical benefits of circumcision have been a topic of debate over the past 40 years. In 1975, the American Academy of Pediatrics Committee on the Fetus and Newborn stated that there were no valid medical indications for circumcision in the neonatal period.3
In 1985, Wiswell and colleagues reported on a retrospective chart review of infants admitted to the hospital with a urinary tract infec-tion in the first year of life.4 They identified a siginificant association between circumcision status and the risk of UTI in boys. A non-circumcised male was 20 times as likely to have been admitted for a UTI as a circumcised boy. This finding and subsequent work done by Wiswell and Roscelli prompted the Academy of Pediatrics to take another look at the circumcision issue. In 1989, the Task Force on Circumcision issued a new statement noting that circumcision had potential medical benefits and advantages as well as disadvantages and risks.6
This new statement implies a need to weigh the benefits against the risks prior to choosing circumcision for a male infant. However, this task is difficult in the absence of quantitative estimates of the likelihood of positive and negative outcomes and the parents values regarding these outcomes.
Decision analysis is a tool that allows one to consider explicitly the uncertainties involved in a given decision and to place values on its eventual outcomes.7 Expected utilities can then be calculated for competing choices, with the decision maker choosing the strategy of highest anticipated benefit. Sensitivity analysis allows one to consider the effect of possible error in the assignment of values to uncertain parameters, and shows how varying these assigned values might change the eventual decision.
A decision model was therefore constructed to address the following two key questions: 1) For a set of values that a reasonable parent might assign to the potential outcomes, using the data of Wiswell and others to assign probabilities, what is the circumcision choice of highest expected benefit? and 2) How does the rate of UTI in the first year of life for uncircumcised males affect the decision?
The model assumes that renal pathology is the single significant untoward outcome of primary urinary tract infections. Wiswell and Geschke8 found in a group of 35,929 uncircumcised male infants that non-circumcision was associated with death from invasive bacterial disease in two male infants with UTI in the first month of life. However, their data do not address the issue of causality, and neonatal UTI is generally thought to be acquired through hematogenous spread.9
A second assumption of the model is that a higher rate of urinary tract infection is the only significant untoward outcome of non-circumcision in the newborn period. Penile cancer occurs almost exclusively in the uncircumcised male at a rate of eight per million.10 The low frequency of this event and the fact that a male would have a "second chance" as an adult to choose circumcision and avoid this outcome may make it reasonable to delete it from the model. Sensitivity analysis will be used to assess the effect of this assumption.
The model does not directly address the issue of sexually transmitted diseases and circumcision due to the absence of prospective studies that control for confounding influences and to the conflicting evidence from retrospective data.11,12 While there are retrospective studies suggesting a higher risk of penile inflammatory conditions, such as balanitis, in uncircumcised males, a recent prospective study from New Zealand was unable to find a statistically significant higher risk of penile problems in uncircumcised boys in the first eight years of life.13 Sensitivity analysis allows one to see the potential effect of ignoring these issues in the model.
The model begins at a choice node or point of decision and goes through chance nodes where the uncertainties of events following the decision are represented. After each final branch, there is a box representing a specific outcome.
The model is displayed in Figure 1. The choice of having or not having a circumcision is represented at the origin of the model. If the boy is circumcised, either things will go smoothly or there will be complications. In a review by Kaplan,14 the likelihood of minor complications, including bleeding, errors of omission and commission in tissue removal, skin bridge, infection, and meatitis, have ranged from 0.1% to 35%. Even though penile denudation and death from sepsis as a result of circumcision have been reported. by all accounts, major complications are exceedingly rare and have not been included formally in the model.15,16 The probability of minor complications was set at 21.8%. Through sensitivity analysis one can see the effect of this potential overestimation on the preferred choice. A boy with a complication of circumcision will either have a urinary tract infection in the first year of life or he will not. The probability of a UTI in the first year of life was taken from the data of Wiswell et al4 and set at 0.2%.
A circumcised male with complications and a UTI may have renal scarring or escape this complication. The likelihood of renal scaring was taken from the work of Winberg et al,17 and set at 7.5%.
The outcome box labeled Kidney scar, Complication, Pain represents a circumcised boy who has had some complication and a urinary tract infection that left him with renal scarring. The second outcome, labelled Infection, Complication. Pain," represents a circumcised boy who has had a complication and a UTI without scarring.
Moving back to the tree branch representing a circumcised male with a complication but no urinary tract infection, the outcome represents the pain of the surgery and the complication.
A boy who has had a circumcision without complication may or may not have a urinary tract infection in the first year of life. Once again, the data of Wiswell et al4 are used for the probability assignment.
The UTI may or may not lead to scarring. These data of Winberg et al17 are again used for this probability.
Then the next outcome state, therefore, represents a boy who has suffered the procedure without complications and has had urinary tract infection leading to renal scarring. The outcome box labelled "Infection, Pain" represents the circumcised boy without complications who has had a UTI without scarring.
A circumcised boy without any complications may alternatively have no urinary tract infection. Therefore. the only price paid is that of the pain associated with the procedure.
Returning to the choice node of the model, a boy who is not circumcised may or may not have a urinary tract infection in the first year of life. The data of Wiswell et al4 are used to set the probability of urinary tract infection in the first year of life at 4.1%.
A non-circumcised boy having a urinary tract infection may or may not have a kidney scar. Once again, the probabilities of Winberg et al17 are used A non-circumcised boy having a urinary tract infection leading to a kidney scar ("kidney scar") will not have experienced the pain of the circumcision "Infection" represents the non-circumcised male with a UTI but no scarring. A non-circumcised male may escape a urinary tract infection. This outcome block is labeled "normal."
The values of the hypothetical "reasonable parent" were set on a linear scale from 0 to 1, beginning with the worst outcome and ascending to the best possible outcome (Table 1). It should be noted that sets of parents will differ in their assignment of values to the outcomes. Sensitivity analysis will show the effect of manipulating some of these values on the preferred choice.
THE OUTCOME STATES AND THEIR ASSIGNED VALUES Outcomes Values Normal 1.00 Pain 0.95 Complication, Pain 0.85 Infection 0.70 Infection, Pain 0.65 Infection, Complication, Pain 0.55 Renal scar 0.15 Renal scar, Pain 0.10 Renal scar, Complication, Pain 0.00
The expected benefit of the two choices can be calculated for the set of assigned values through the process known as "averaging out and folding back the tree."7 This process starts at the outcome points and works back to the root of the tree. The value associated with each outcome is multiplied by the probability of that outcome. These averaged values of each possible outcome of a chance node are then added together. The sum is then multiplied by the probability of its occurrence, which is taken from the branch of the tree that precedes it. The process continues until the decision node is reached. The clinician then chooses the strategy of the highest anticipated benefit. Table 2 displays the calculated expected utilities. The choice of no circumcision is preferred because it give a higher expected benefit than the choice of circumcision.
Table 2 EXPECTED BENEFITS Expected Utilities (Benefit) Circumcision 0.9275 No Circumcision 0.9860 The expected benefits (maximum = 1) of the two choices as calculated in the model
Table 3 shows the effect of some of the assumptions on the preferred choice generated by the model. Unless the likelihood of a urinary tract infection in a non-circumcised male reaches 29%, the preferred choice remains non-circumcision. This value of 29% is defined as the "threshold" because it represents the value above which the preferred decision would change. The rate of minor complications associated with the procedure has no effect on the preferred choice. No threshold is reached, and the choice remains non-circumcision even if no babies have minor complications from the procedure, The choice of highest expected utility is non-circumcision for all probabilities of minor complications in this model.
SENSITIVITY ANALYSIS Variable Value Threshold pUTInoCirc 0.041 0.29 pComplications 0.218 "Normal" 1.0 0.9098 Pain 0.90 0.9867 Sensitivity analysis of 4 variables in the model pUTInoCirc = the probability that a non-circumcised infant will have a UTI in the first year. pComplications = the probability of minor complications from a circumcision.
The model assumes that the only untoward side effect of noncircumcision is the risk of urinary tract infections. If there were other significant risks of non-circumcision, then the value of the "normal" state would be discounted. When the value of the normal state drops below 0.9098, the preferred choice becomes circumcision. This could occur if it were shown that the risk of sexually transmitted diseases and their side effects was significantly greater in non-circumcised males than in circumcised males. This would also apply if the rate of penile cancer was significantly higher and it was shown that circumcision later in life did nothing to prevent this outcome.
Sensitivity analysis also demonstrates that parents who consider the pain associated with the procedure to be of trivial concern will choose circumcision. The decision is reached when the pain state is valued at 0.9867 or higher.
For the set of values assigned to the possible outcomes, the choice of no circumcision yielded the highest expected benefit. The preferred choice would remain no circumcision given this set of values, unless the risk of urinary tract infection in the first year of life was at least 29% or greater.
Sensitivity analysis suggests that the decision of whether or not to circumcise a boy hinges more on the parents' values regarding the pain of the procedure than on the risk of UTI. Parents who discount this pain as a short-lived discomfort may choose circumcision, while others who are bothered by the notion of surgery without anesthesia would opt against the procedure.
Recommending circumcision for all newborn boys to avoid penile carcinoma ignores the possibility of alternative strategies to lower this risk and the mature male having the opportunity later in life to decide to have a circumcision with anesthesia. Data showing the failure of other strategies to lower the risk of penile carcinoma would lower the expected benefit of non-circumcision. Safe and effective means of anesthesia for newborn circumcision may also significantly affect the decision for many parents. The current concern regarding the AIDS epidemic and the suggestion that circumcision may reduce the risk of this disease18-19 has prompted further calls for circumcision for all boys. While the risk of AIDS attributable to non-circumcision is unknown, it is unlikely that it would be large enough that circumcision as a primary prevention strategy would have much effect.
Decision analysis should not be viewed as a means of escape from difficult choices. Rather, it is a tool that forces the decision makers to consider explicitly the uncertainties involved, to see the effects of potential errors in these estimates, and to assign values to the outcomes of their decisions.
The choice of circumcision, excluding those for religious beliefs or cultural reasons, must be made by well-informed parents and should not be dictated by the risk of urinary tract infection alone.