Although there is widespread agreement that physicians and policymakers should use all reasonable means to reduce the risk for transmission of HIV to patients, the best approach to reducing this risk is unclear. A recent analysis [10] and retrospective studies that have interviewed and tested patients treated by health care providers infected with HIV indicate that the chance for transmission of HIV from provider to patient is small. By the end of 1993, approximately 22 000 patients who had been treated by infected providers had been examined; these examinations found no evidence that HIV had been transmitted during an invasive procedure. Despite these studies, concern remains about the possibility of transmission of HIV from providers to patients. Because of growing evidence that such transmission is rare and because a program to screen surgeons and prohibit infected surgeons from doing invasive procedures is potentially expensive, we assessed whether the benefits of such a program would justify its costs.

Our analysis considers the potential benefits of a screening program and the costs associated with screening, early treatment, and a policy that prohibits infected surgeons from doing invasive procedures. In contrast to previous studies, our analysis expresses the benefit of the screening program in terms of both the number of HIV infections prevented and the years of life saved by preventing these infections. Estimation of the years of life saved allows the cost-effectiveness of a surgeon-screening program to be compared with the cost-effectiveness of other health interventions. Because a program to screen surgeons may benefit the surgeons as well as their patients, we defined the health benefits of the program comprehensively to include not only the benefit to patients (from fewer transmissions of HIV) but also the benefit to surgeons (from early medical intervention for those identified as having HIV infection) and to the sexual partners of identified surgeons (from fewer transmissions of HIV). Failure to include these additional benefits would underestimate the value of a screening program. Throughout the analysis, therefore, our estimates of cost-effectiveness incorporated all of these benefits. We also examined the ways in which the cost-effectiveness of a screening program for HIV would be affected by changes in two critical variables: the prevalence of HIV infection among surgeons and the likelihood that an infected surgeon will transmit HIV to a patient during an invasive procedure.

Methods

Top Methods Results Discussion Author & Article Info References

We did a cost-effectiveness analysis of a program to screen surgeons and to prohibit surgeons with HIV infection from doing invasive procedures. To calculate the years of life that would be saved by a screening program for HIV infection, we used a decision model to estimate both the number of infections that would be prevented by such a program and the years of life that would be saved by preventing HIV infections. We calculated the dollar expenditure required to save these years of life by estimating the costs of the program.

Defining Benefit

A screening program for HIV infection may benefit both the person identified as having HIV infection (who can receive early medical intervention) and the public at large (if the person identified as infected reduces high-risk behavior that may transmit infection, such as sexual contact or needle-sharing). A program directed at surgeons may also result in fewer transmissions of HIV to patients; this benefit is the primary emphasis of our analysis. To avoid underestimating the overall health benefit of a screening program for HIV infection, our analysis considers the benefits to the surgeon identified as having HIV infection, to the sexual partners of that surgeon, and to the patients treated by that surgeon.

Analytic Model

We used a decision model to do the analysis (Figure 1). This model accounts for each of the benefits of the screening program and for the costs incurred; it is a 32-state Markov cohort described more fully in the Appendix. We used a statistical model based on data from the San Francisco City Clinic Cohort [11] to estimate length of life in patients with HIV infection [10, 12]. We calculated the years of life saved when an HIV infection is prevented as the difference between life expectancy with and without HIV infection. To estimate the effect of early identification through screening on quality of life, we used quality adjustments derived from a survey of 128 physicians [13] for the following states of health: asymptomatic HIV infection, symptomatic HIV infection without AIDS, and AIDS.

View larger version (21K): [in this window] [in a new window]   Figure 1. Schematic representation of the decision model. The square node at left represents the decision to screen. If a physician is screened, the test result is either positive or negative. For each of these results, the decision model calculates the probability of HIV infection by using the sensitivity and specificity of the HIV screening tests and Bayes theorem. The events after screening are modeled in a 23-state Markov model (see Appendix). For persons who are not screened, a 9-state natural history Markov model simulates the natural history for physicians infected with HIV and estimates the age-specific life expectancy of those physicians without HIV infection.

The data used in the model [11, 12, 14-71] are shown in Table 1. We used data taken from the literature [46, 54-63] to estimate the effect of early medical intervention on length of life. We consider early medical intervention to include the following currently recommended therapies: treatment with antiretroviral medications, prophylaxis against Pneumocystis carinii pneumonia and other opportunistic infections, and screening and treatment of coexisting tuberculosis and sexually transmitted diseases [78-80]. In our base-case analysis, we assumed that early medical intervention increases length of life with HIV infection by approximately 1.0 year; we explored the effect of this assumption in sensitivity analyses. Health benefits and costs were discounted at 5%. The perspective of the analysis was that of society.

To analyze the costs and benefits of a screening program, we assumed that only surgeons identified through the mandatory screening program would receive early medical intervention, that no screening would occur without a mandatory screening program, and that the policy to restrict infected surgeons from doing invasive procedures would be 100% effective. Each of these assumptions created a bias in favor of a screening program. In fact, many surgeons with HIV infection would probably be identified and receive early medical intervention in the absence of a mandatory screening program [22, 81-83]. These surgeons may also restrict their practices; by excluding this possibility, we attributed more benefit to a mandatory screening program than may actually exist. We examined the effect of these assumptions in sensitivity analyses.

Estimating the Probability of HIV Transmission

The likelihood that HIV transmission will occur during an invasive procedure depends on the probability that a patient will be exposed to a contaminated instrument and the probability that a patient will seroconvert if so exposed. We based our estimates of the probability that a patient will be exposed during a procedure on modifications of a model developed by the CDC [23] (see Appendix). To estimate the likelihood that a patient will become infected after a percutaneous exposure, we used two lines of evidence. First, in 63 studies that evaluated 22 032 cases of patients treated by providers infected with HIV [84], no cases of transmission of HIV from provider to patient were documented. Although these studies suggest that transmission of HIV to patients occurs rarely, it is not possible to infer directly from these studies the probability of transmission to a patient who sustains a percutaneous exposure. It is not known how many patients treated by the providers in these studies actually sustained percutaneous exposures to instruments contaminated with infected blood. Second, in a summary of 13 prospective studies of transmission of HIV to health care workers who were exposed to the blood of infected patients through needle-stick injuries [27], the probability for the transmission of HIV was 0.29%. Most of these injuries were single needle-sticks from hollow-bore needles. We used 0.29% as the probability of transmission in our base-case analyses. In sensitivity analyses, however, we analyzed the effect of seroconversion rates after recontact with a contaminated instrument that are up to 20 times higher than those observed in health care workers who had seroconversion after receiving needle-stick injuries.

To account for the possibility that some surgeons might transmit HIV more frequently than other surgeons (resulting in clusters of transmission similar to those seen with the hepatitis B virus [85, 86]), we also analyzed the effect screening would have if 5% of surgeons were to transmit infection at a rate consistent with the rate observed in the dental practice in Florida [87, 88]. The dentist in Florida transmitted HIV to 6 of approximately 1100 patients in his practice who were tested. By assuming that his patients had had an average of 2 invasive procedures, we estimated that transmission had occurred during 6 of 2200 procedures [82]. An increase in our base-case estimate of the probability of transmission of HIV after percutaneous exposure to the patient (0.29%) by 118 times provided transmission rates consistent with those observed in the dental practice.

Screening and Risk Behaviors

To estimate the effect of a screening program on the transmission of HIV to sexual partners, an estimate of the effectiveness of counseling in changing high-risk sexual behavior is needed. The effect of screening, testing, and counseling on subsequent risk behaviors has been studied in various populations [38-4589, 90]. The degree of behavioral change varies from study to study [38] but, in general, only modest changes in behavior have been observed. In addition, secular trends in risk behavior (such as the decrease in risky sex in populations of homosexual men) confound many studies, making it difficult to assess the degree to which knowledge of HIV status alone determines changes in risk behavior. Because of the modest but variable effects seen in these studies, we assumed for our base-case analysis that counseling and testing results in a 15% decrease in risk behaviors (for example, a patient would decrease his or her number of sexual partners by 15% or would intermittently use condoms to reduce infectivity by approximately 15%) in persons with positive HIV test results. We explored a range of values in sensitivity analyses.

Policy to Restrict Activities of Physicians with HIV Infection

In our primary analysis, we evaluated the effect of a program that screens surgeons for HIV and that restricts surgeons with HIV from doing invasive procedures but allows them to continue to do other types of patient care. We focused on surgeons in our primary analyses because they do more invasive procedures than other physicians. A screening program that includes all physicians would at best be equally cost-effective and would probably be substantially less cost-effective than a program to screen only surgeons.

One consequence of a program to screen surgeons would be a decrease in services provided by the surgeons identified as infected with HIV; this decrease in service represents an induced cost of the program. Although there is no ideal way to quantify such a cost, we used the change in income of the surgeons as a surrogate for that cost. In our base-case analysis, we assumed that the income of a surgeon identified as infected with HIV would decrease from the average annual salary of a surgeon ($251 000) to the average annual salary of an internist ($168 500) [73]; thus, for each surgeon identified as infected, we attributed an annual induced cost of approximately $83 000 to the policy. This cost would be partly offset if, as we assumed, surgeons who are identified and who receive early medical therapy remain well and able to work approximately 1 year longer than they would if they were not identified. This assumption creates a bias in favor of screening because some surgeons would probably not be able to work for a full additional year. Finally, although we believe that the induced costs of the screening program should be included in an economic analysis, an alternative approach is to consider only direct costs in the analysis. To evaluate the effect of the latter approach, we evaluated the cost-effectiveness of screening both with and without induced costs.

Costs

Our analysis included the direct costs of screening and treatment, the induced costs of preventing surgeons infected with HIV from doing invasive procedures (except where noted), and the cost savings associated with the prevention of infections in sexual partners and patients (Table 1). We estimated the direct costs of screening (testing and counseling) on the basis of a survey of publicly funded test centers in California (Owens DK. Unpublished data, 1993) and on a study sponsored by the CDC [74]. The cost of treatment was based on estimates by Hellinger from the ACSUS (AIDS Cost and Service Utilization Survey) study [64, 65]. Further details of cost estimates are presented in Table 1.

Results

Top Methods Results Discussion Author & Article Info References

The estimated health and economic outcomes of a screening program are shown in Table 2. A one-time national screening program would prevent approximately 4.3 infections (range, 1.9 to 21.3 infections) among patients having more than 180 million surgical procedures. Prevention of these infections would result in approximately 21 additional years of life for these patients as a group. The screening program would identify 137 surgeons infected with HIV (among 141 646 surgeons screened); we estimate that the 137 surgeons, as a group, would gain approximately 69.4 additional years of life due to early medical intervention (after discounting, see Table 2. An additional 0.9 infections [range, 0 to 12.9 infections] would be prevented in sexual partners of the identified surgeons; these partners would gain an additional 28.3 years of life. Prevention of an HIV infection in a patient or sexual partner would result in a modest economic saving (Table 2).

Cost-Effectiveness of Screening

The marginal cost-effectiveness of the screening program, calculated by including both direct and induced costs, is shown in Table 3; it is calculated relative to the alternative of not screening surgeons. Included are all benefits that accrue from screening: fewer transmissions to patients and sexual partners and the increased length of life resulting from early medical intervention. Table 3 shows the cost per year of life saved, calculated on the basis of analyses that include only the mortality resulting from HIV infection. Also shown is the cost per quality-adjusted year of life saved, calculated on the basis of analyses that include the effect of HIV infection on both length and quality of life. A one-time screening program results in expenditures of $458 000 per year of life saved (range, $147 000 to $687 000 per year of life saved).

When the morbidity of HIV infection is considered, a screening program is less cost-effective Table 3 because it identifies and "labels" surgeons as HIV infected before they would otherwise have been identified. Relative to good health, there is a decrement in quality of life associated with having asymptomatic HIV infection, a decrement that partially offsets the increase in length of life associated with early medical intervention.

A screening program designed to prevent transmission of HIV to patients would probably require periodic rather than one-time screening. Figure 2 indicates how the costs and benefits of screening change as the screening interval is changed. As the frequency of screening increases, screening becomes less cost-effective. If surgeons are screened every 10 years, the program results in expenditures of $597 000 per year of life saved (range, $187 000 to $973 000 per year of life saved). Screening at 7- and 4-year intervals results in expenditures of $632 000 and $699 000 per year of life saved, respectively. Annual screening results in expenditures of approximately $1.1 million per year of life saved (range, $338 000 to $1 886 000 per year of life saved).

View larger version (13K): [in this window] [in a new window]   Figure 2. Effect of changes in screening frequency on the cost and benefits of screening. One-time screening results in the most favorable cost-effectiveness ratio; annual screening results in the least favorable cost-effectiveness ratio.

If the induced costs of the screening program are excluded, the estimated cost-effectiveness of screening improves substantially (Table 4). A one-time screening program requires expenditures of $91 000 per year of life saved and $298 000 per quality-adjusted year of life saved. Annual screening requires expenditures of $539 000 per year of life saved. When the morbidity of HIV infection is included in the analysis, screening is substantially less cost-effective.

Screening Physicians Other Than Surgeons

To examine the effect of screening physicians other than surgeons, we analyzed the cost-effectiveness of screening for a population in which the number of invasive procedures done annually by the infected physician was reduced to 50, the induced costs per infected physician identified were reduced to $10 000, and the probability of percutaneous exposure to the physicians was 2.5% (unrealistically high for nonsurgical procedures, creating a bias in favor of screening). Given these assumptions, an annual screening program would cost $591 000 per year of life saved.

Sensitivity Analyses

We examined all variables of the model in sensitivity analyses. The sensitivity analyses that we report are based on analyses of the cost-effectiveness of an annual screening program unless otherwise noted, and they incorporate only the effect of the screening program on length of life. The results were qualitatively similar when morbidity was also included in the analysis, but in each case the program was less cost-effective (data not shown).

Sensitivity analyses done on model variables do not alter the main conclusions of our analysis. Figure 3 shows the effect on the cost-effectiveness of screening of variation in the probability of seroconversion after a patient sustains a percutaneous injury with a contaminated instrument. The cost of an annual screening program does not decrease to $75 000 per year of life saved until the probability of seroconversion after patient exposure reaches approximately 16.5% (60 times our base-case estimate). Figure 4 indicates the effect of the prevalence of HIV infection among surgeons on the cost-effectiveness of screening. With our base-case estimate of the induced costs of the program included ($82 551 annually per infected surgeon identified), the cost of annual screening remains more than $500 000 per year of life saved regardless of the prevalence of HIV infection among surgeons. This surprising result occurs because, although the benefit from a screening program increases as the prevalence of HIV infection increases, the induced cost of the program increases also, preventing the cost-effectiveness ratio from decreasing. If counseling induced larger reductions in risk behavior than assumed in our base-case estimate, screening becomes more cost-effective. For example, the costs of an annual screening program are $728 000 and $487 000 when the reduction in risk behavior is 50% and 100%, respectively. Sensitivity analyses also indicate that doubling the effectiveness of early medical intervention (extending length of life by 2 years) or increasing the number of sexual partners at risk to four does not alter our conclusions.

View larger version (29K): [in this window] [in a new window]   Figure 3. Effect of changes in the probability of seroconversion after a patient is exposed to a contaminated instrument on the cost-effectiveness of an annual screening program. As the probability of seroconversion after exposure increases, the screening program becomes more cost-effective. The curves represent the cost-effectiveness as estimated with both direct and induced costs (top curve) and with direct costs only (bottom curve).

View larger version (25K): [in this window] [in a new window]   Figure 4. Effect of changes in the prevalence of HIV infection among physicians, including all benefits and costs, on the cost-effectiveness of an annual screening program. The curves represent the cost-effectiveness as estimated with both direct and induced costs (top curve) and with direct costs only (bottom curve).

As noted above (see Methods), many surgeons may have been screened voluntarily in the absence of a mandatory screening program. If these surgeons provided proof of their HIV test results, they would not necessarily be rescreened in a one-time screening program. For example, if 50% of surgeons have been screened voluntarily, the direct costs of mandatory screening would be reduced by 50%. We reanalyzed the cost-effectiveness of a mandatory screening program by excluding both the direct costs associated with screening 50% of surgeons and the benefits to these surgeons (from early medical intervention) and to their sexual partners (from fewer transmissions of HIV). For those surgeons screened voluntarily, it would be inappropriate to attribute to the mandatory screening program the benefit they or their sexual partners receive from screening. Under these circumstances, expenditures for a one-time mandatory screening program increase from $458 000 to $699 000 per year of life saved because the percentage decrease in benefits is greater than the percentage decrease in costs. This is true with one exception: If induced costs are excluded, voluntary screening of 50% of surgeons makes a mandatory program more cost-effective ($76 000 compared with $91 000 per year of life saved).

We also analyzed voluntary screening programs. Strictly voluntary screening programs would be as cost-effective as the mandatory program we analyzed if surgeons identified as having HIV infection completely restricted their practices to noninvasive care; they would be less cost-effective if compliance was less than 100% (data not shown). This result occurs because our base-case analysis did not attribute a cost for enforcement of a mandatory program.

To examine the influence of clustered transmission on the cost-effectiveness of screening, we assumed that 5% of surgeons would transmit HIV infection at approximately 118 times the base-case rate (consistent with the rates observed in the dental practice in Florida). Under these circumstances, one-time and annual screening programs would result in expenditures of $220 000 and $466 000 per year of life saved, respectively. The cost of one-time screening does not decrease to $75 000 per year of life saved until 23% of infected surgeons transmit HIV at 118 times our base-case rate, and the cost of annual screening reaches $75 000 per year of life saved only when 48% of infected surgeons transmit HIV at rates consistent with that of the dentist in Florida.

The scenario most favorable to screening is a one-time screening program analyzed using the assumptions that no induced costs result from preventing surgeons from operating and that the prevalence of HIV infection among surgeons is higher than was estimated in our base-case analysis. Given these assumptions, the cost of a one-time screening program is approximately $56 000 per year of life saved when the prevalence of HIV infection among screened surgeons is 0.2% (two times the base-case estimate); it decreases to $39 000 per year of life saved when prevalence is estimated at 0.4%. The cost of an annual screening program, however, does not decrease to $75 000 per year of life saved until the prevalence of HIV infection in screened surgeons is approximately 1.9%.

The annual induced cost from restriction of a surgeon's practice has a marked effect on the cost-effectiveness of a screening program. The cost of an annual screening program is $443 000 per year of life saved when the annual induced cost is $5000; $739 000 per year of life saved when the annual induced cost is $40 000; and $2 526 000 per year of life saved when the annual induced cost is $251 000. This last figure, $251 000, is the average annual salary of surgeons, including obstetricians and gynecologists.

Discussion

Top Methods Results Discussion Author & Article Info References

Transmission of HIV to six patients in a Florida dental practice raised the question, both within the medical profession and among the public, of whether invasive procedures done by providers infected with HIV pose an unacceptably high risk to patients. Given the information currently available, our analysis indicates that screening surgeons for HIV infection to prevent transmission to patients would result in expenditures per year of life saved that are considerably in excess of those of most accepted health interventions. We estimate that a one-time national screening program of surgeons would prevent 1.9 to 21.3 HIV infections in patients in the United States. The costs of the program would include approximately $8 million for screening and incremental medical treatment and approximately $44 million for the decrease in services provided by surgeons identified as HIV infected. We estimate the marginal cost-effectiveness of a one-time screening program of surgeons to be approximately $458 000 per year of life saved when we include both direct and induced costs. In comparison, screening for hypertension results in expenditures of $12 200 to $42 600 per year of life saved (expressed in 1993 dollars [91]). Thus, the program is expensive relative to other health interventions even though our analysis incorporates the potential benefit to surgeons and their sexual partners in addition to the benefit to patients.

If screening were done annually or biannually, as would be probable in any ongoing program, it would be considerably less cost-effective because the first episode of screening would detect almost all prevalent cases of HIV infection. In subsequent screening episodes, most detectable cases of HIV infection would be those that had occurred since the previous screening episode. Because the available evidence, despite being limited, suggests that incidence of HIV infection in surgeons is low, few additional cases of HIV infection would be detected. Thus, as the interval between screening tests shortens, the screening program becomes less cost-effective. Screening at intervals of less than 1 year would cost more than $1.1 million per year of life saved.

Our primary analyses evaluated the cost-effectiveness of a program to screen surgeons. A program designed to screen all physicians would be less cost-effective than one targeted to surgeons because other physicians do fewer invasive procedures. With fewer opportunities to transmit HIV to patients, the benefit of the program would diminish but the associated costs would not be proportionally lower.

Several limitations of the data used in our study deserve comment. Central to our evaluation is our estimate of the probability that a surgeon infected with HIV will transmit HIV to a patient. As noted, only indirect evidence is available to estimate the probability that a patient will become infected if he or she is exposed to an instrument contaminated by infected blood. Studies of needle-stick exposures [27] and retrospective studies suggest that the likelihood of transmission of HIV during an invasive procedure is remote. However, the inoculum a patient receives during a procedure done with a contaminated instrument may be higher than that from a single needle-stick exposure, in part because if the contamination of a surgical instrument is unrecognized, the instrument may be used repeatedly. In addition, although 22 000 cases have been investigated in retrospective studies [84], the number of patients actually exposed to contaminated blood [24] is undoubtedly much smaller (estimated as 22 000 x 2.5% [the average exposure rate for the surgeon] x 32% (the recontact rate) 175 patients), which indicates that these studies did not exclude low but nonzero transmission rates.

A related issue is whether transmission of HIV from provider to patient is endemic or whether it occurs in outbreaks [92]. In our base-case analysis, we assumed that transmission was endemic (that each infected provider was equally likely to transmit infection). If transmission from surgeon to patient occurs in outbreaks or clusters (if a particular provider is more likely than others to transmit HIV), our analysis may underestimate risk. In sensitivity analyses, however, we found that if 5% of surgeons transmitted at approximately 120 times the base-case rate (consistent with the rate observed in the Florida dental practice), mandatory screening was still expensive. Given the current evidence, including investigation of more than 60 health care workers with HIV infection, it is unlikely that the clustered transmission would occur with greater frequency than that which we examined in sensitivity analyses; thus, the conclusions of our analysis remain unchanged for plausible scenarios about the mechanism of transmission. If new evidence of clustered transmission becomes available, however, the cost-effectiveness of strategies to identify the providers should be reevaluated.

In addition, the prevalence of HIV infection among surgeons is not precisely known. Our initial analyses assumed that the prevalence and incidence of HIV infection in surgeons were similar to those of physicians in the military. Because physicians in the military know that they will be screened for HIV, self-selection may render these studies unrepresentative of physicians in other environments. Although the prevalence of HIV among surgeons is likely to vary among clinical settings (for example, the prevalence among surgeons may be higher in geographic locations in which the prevalence of HIV among patients is higher), our sensitivity analyses indicate that periodic screening would be expensive even if the prevalence of HIV infection among surgeons is 1%, a value that is unrealistically high. The prevalence of HIV infection becomes important only if the costs of prohibiting surgeons from doing invasive procedures are excluded completely.

Thus, we believe that the conclusions of our analysis are robust even though further information on the likelihood of transmission during invasive procedures and on the seroprevalence of HIV among physicians would be useful. Our sensitivity analyses indicate that the uncertainty in these variables is not sufficient to alter our conclusions about annual screening programs.

An important determinant of the cost-effectiveness of screening is the induced cost of preventing a surgeon from doing invasive procedures (Tables 3 and 4). We estimated the value of the lost services of surgeons who are prevented from doing invasive procedures as equal to the difference in income between a surgeon and an internist. Because some surgeons identified as having HIV infection would probably stop practicing completely (resulting in a greater loss of services) and because we assumed that surgeons who are identified through screening will be able to work an additional year because of early medical intervention, our estimate of the cost may be too low [93]. Some have argued, however, that surgeons create demand for their services [28] and that a decrease in the number of surgeons would be beneficial. We view this as a separate policy issue and believe that it would be inappropriate to exclude the value of surgeons' lost services from the evaluation of a policy that is designed to prevent transmission of HIV. If these costs are excluded Table 4, screening is substantially more cost-effective, although annual screening is still expensive relative to the benefit obtained.

Other investigations of the costs and benefits of screening providers have analyzed different populations and policies than we evaluated. In addition, the authors of these studies used the number of HIV infections prevented or the number of infected physicians identified as primary end points; they did not directly estimate the cost per year of life saved. Gerberding [94] evaluated the costs associated with a program to screen all providers (including physicians, dentists, nurses, and paramedics) and to prohibit infected providers from doing invasive procedures at a particular hospital. This analysis estimated a cost of approximately $287 000 per health care worker identified during the first year alone. The cost of restricting providers' practices contributed more than 50% of the estimated expenditures. An analysis of annual screening that did not include induced costs of practice restriction estimated costs of more than $50 million per HIV infection prevented [95]. Chavey and colleagues [96] estimated expenditures of $9 177 615 per transmission prevented in their hospital. Their analysis did not consider benefit to the screened surgeons or their sexual partners and included direct costs only. They concluded that the cost-effectiveness ratio for screening was much greater than that of other accepted health interventions. Sell and colleagues [88] concluded that neither voluntary nor mandatory screening was "convincingly" cost-effective. In contrast to our analysis, they found mandatory screening programs to be more expensive than voluntary programs. Our results differ because these investigators did not consider the induced cost of screening and because they assumed that 90% of infected physicians would be screened in a voluntary program and that 90% of those identified would restrict their practice appropriately; these optimistic assumptions make voluntary screening appear more efficient.

In a comprehensive analysis of policy options for screening physicians and dentists, Phillips and colleagues [82] estimated that the value of a prevented HIV infection would have to exceed $20 million for the benefits of screening to outweigh the costs when productivity losses from surgeons are included in the analysis. These authors also found, however, that estimates of the costs and benefits of screening physicians for HIV varied substantially depending on model assumptions. They concluded that mandatory screening programs should not be implemented until further evidence of cost-effectiveness was shown. Schulman and colleagues [93] estimated the risk for transmission of HIV during invasive procedures to be low and questioned whether the benefit was justified by the costs of the program. In summary, no studies have concluded that the benefits of a program to screen physicians justify the costs incurred. Our study differs from previous analyses in that we considered the benefit of screening to surgeons and their partners and the effect of screening on quality of life, and in that we estimated the expenditures required per year of life saved rather than per infection prevented. This metric enabled us to compare HIV screening with other health interventions.

Is screening surgeons for HIV cost-effective? The choice of a particular cost-effectiveness threshold is controversial [97-100]. Most commonly accepted interventions, however, cost between $10 000 and $150 000 per year of life saved [91, 101-103]. Our estimates of the cost of screening surgeons for HIV exceed this range substantially. Thus, our evaluation indicates that a program to screen surgeons to prevent transmission of HIV to patients is an unwise use of public resources, given what is currently known about the epidemiology of transmission of HIV. Estimates of the chance of transmission during an invasive procedure rely on indirect evidence, however, and we therefore recommend that current retrospective studies be continued. Our analysis also indicates that the effect of early identification and treatment of HIV infection on quality of life is an important, although incompletely studied, determinant of the cost-effectiveness of screening for HIV. The effect of early identification on quality of life is relevant to HIV screening efforts for both patients and providers and warrants further investigation. Implementation of screening programs for physicians for the purpose of preventing transmission of HIV to patients should be postponed until the time, if ever, when surveillance studies indicate that transmission rates of HIV during invasive procedures are substantially higher than current evidence suggests.

Appendix

The decision model used in our cost-effectiveness analysis compares the costs and benefits of a screening program for HIV infection with the status quo, in which no screening is done. The screening strategy is represented by a 23-state Markov model [104]; the "no-screening" strategy is represented by a 9-state Markov model. Each state incorporates natural history (no HIV infection, asymptomatic HIV infection, symptomatic HIV infection without AIDS, AIDS, and death), screening status (screened positive for HIV infection, screened negative for HIV infection, or unknown), and type of infection (prevalent or incident). Persons start in the model as unscreened and either uninfected or infected. Persons who are screened are identified as infected or uninfected on the basis of the sensitivity and specificity of the HIV tests (Table 1). As time proceeds, persons move among the states. Transitions occur along two dimensions: Disease state can progress (for example, from symptomatic infection to AIDS) and information state can change (for example, from being unscreened to being screened positive for HIV infection). We modified transition probabilities for the natural history of HIV infection, originally determined by Longini and colleagues [11], to include age-specific mortality from all causes. The mean duration of health states predicted by the model are consistent with epidemiologic studies of the natural history of HIV infection [105-107]. To calculate quality-adjusted years of life saved, we applied quality adjustments to each of the health states associated with HIV infection. The model analyzes one-time and sequential screening. Costs and health benefits are discounted at an annual rate of 5%.

Our estimate of the likelihood of transmission of HIV during an invasive procedure is based on modifications of a model developed by the CDC [23]. For a physician to transmit HIV to a patient during an invasive procedure, the physician must sustain a percutaneous injury that contaminates a needle or surgical instrument, the contaminated instrument must recontact the patient (which constitutes a percutaneous injury to the patient), and, finally, the patient must seroconvert [23]. Thus, the probability of acquiring HIV during a particular procedure is the product of the percutaneous exposure rate, the recontact rate, and the seroconversion rate given an exposure. In a prospective, observational study of 1382 surgical procedures [22], the probability that the surgeon would receive a percutaneous injury averaged 2.5%; in 32% of these injuries, the contaminated instrument recontacted the patient. We used these values in our base-case analysis (Table 1). This method of estimating the probability of transmission assumes that HIV transmission is endemic. In effect, this approach assumes that all surgeons who have HIV infection are equally likely to transmit HIV. Thus, our base-case calculation does not provide an estimate of the likelihood of transmission that occurs in an outbreak, when a particular surgeon is much more likely than other surgeons to transmit HIV. In sensitivity analyses, we evaluated the cost-effectiveness of screening when a few surgeons transmit at higher rates (clustered transmission).

The health outcomes shown in Table 2 were estimated as follows. The number of surgeons identified by a screening program as having true-positive HIV test results was estimated as the product of the prevalence (0.00097), the number of surgeons who will be screened *RF 141,646 *, and the test sensitivity (0.995). For each person screened, the Markov model calculates the number of infections prevented in patients, the number of years of life gained by patients, the number of infections prevented in sexual partners of surgeons, the number of years of life gained by sexual partners of surgeons, and the number of years of life gained by surgeons because of early medical therapy. The total number for each of these outcomes (for example, the total number of infections prevented in patients; Table 2 was calculated as the product of each of these outputs from the Markov model [expressed per person screened] and the total number of surgeons screened.

The Markov model estimates transmission to sexual partners based on the number of partners at risk, the risk for infection per partnership, and the sex of the infected surgeon (to allow for different rates of male-to-female and female-to-male transmission). The probability of transmission in a sexual relationship is shown in Table 1. The rates used in the analysis assume that male-to-female transmission occurs in 15% of relationships over approximately 4 years and that female-to-male transmission occurs in 3% of partnerships over 3.25 years, as consistent with epidemiologic studies cited in Table 1.

The economic outcomes in Table 2 were estimated as follows. We calculated the direct cost of screening all surgeons as the product of the cost per person screened ($57) and the number of surgeons screened *RF 141,646 *. We calculated the cost associated with early treatment as the product of the number of surgeons with true-positive HIV test results and the net present value of the stream of future medical costs from early treatment. To calculate the costs from early medical treatment, we calculated the discounted difference in the cost for persons who were identified through screening as having HIV infection and the cost for persons who were not screened and thus did not receive early medical therapy. We assumed that screening identifies persons at the midpoint of the asymptomatic period of HIV infection. Thus, persons identified by screening accrued costs for 3.6 years with asymptomatic HIV infection ($5467 per year), for 2.7 years with symptomatic HIV infection but without AIDS ($12 586 per year), and for 2.1 years with AIDS ($35 394 per year). We assumed that persons who were not identified by screening accrued costs for 2.7 years in symptomatic state without AIDS ($11 188 per year) and for 2.1 years with AIDS. As noted in Table 2, for persons who were not identified by screening we estimated lower medical care costs for the period in which the patient had symptomatic HIV infection without AIDS because their HIV status was not known.

The induced cost from loss of surgeons' services was calculated as the net present value of the difference in earnings between a surgeon who is screened and identified and a surgeon who is not screened. We assumed that an infected surgeon who is not screened will work until he or she develops AIDS. A surgeon identified by screening was also assumed to work until he or she developed AIDS, but at the salary of an internist. The calculation assumes that a surgeon identified through screening will receive early medical intervention and will remain healthy enough to work (doing nonsurgical care) for 1 year longer than surgeons not identified through screening. Therefore, the induced cost per surgeon is calculated as the discounted difference between a surgeon who works 6.25 years at $251 000 per year and a surgeon who works 7.25 years at $169 000 per year. We calculated the total induced cost as the product of this number and the total number of surgeons with positive HIV test results, including the estimated 0.8 false-positive test results that would occur in the screening program. We calculated the cost savings from prevention of an HIV infection as explained in Table 2. This calculation assumes that the rate of transmission from an infected surgeon is constant.