ORIGINAL ARTICLES Optimal Management Strategies for HIV-infected Patients Who Present with Cough or Dyspnea= A Cost-Effectiveness Analysis KENNETH A. FREEDBERG, MD, MSc, ANNA N. A. TOSTESON, ScD, DEBORAH J. COTTON, MD, MPH, LEE GOLDMAN, MD, MPH Objective: To determine the effectiveness and costs o f alternative management strategies f o r patients infected with the human i m m u n o d e ~ virus ( H I V ) w h o present with pulmonary symptoms. Design: Decision analysis comparing initial testing (arterial blood gas analysis, induced sputum analysis, or bronchoscopy with bronchoalveolar lavage) with empiric antibiotics (trimethoprim-sulfamethoxazole or erythromycin). Subsequent steps in management are detailed based o n the results o f initial managemenL Patients were stratified by initial CD4 lymphocyte c o u n t (< 2 0 0 / mm 3, 200-500/mm s, or > 500/mm ~) and results o f chest radiography. Setting: Hypothetical Measurements and main results: The estimated levels o f effectiveness among strategies were relatively similar, but costs varied markedly. I f potentially reasonable strategies are defined as those that have incremental costeffectiveness ratios below $50, O00p e r quality-adjusted life year (QALY), the recommended strategies would be: f o r patients at highest risk f o r Pnenmocystls cariniipneumonia (PCP), with a probability o f PCP above 30% (CD4 < 200/mm 3 and almormal chest radiograph orprior history o f PCP), begin with induced sputum analysis ($34,174/ QALY);for intermediate-risk patients, with a probability o f PCP between 6% and 30% (CD4 < 200/mm 3 and normal chest radiograph; or CD4 200-500/mm 3, regardless o f c h e s t radiograph findings), begin with arterial blood gas analysis ($4,593 to $8,310/QALY); f o r low-risk patients, with a probability o f PCP below 6% (CD4 > 500/mn~, regardless o f chest radiograph findings), begin with o n e w e e k o f erythromycin, followed by induced sputum examination i f symptoms persist ($675 to $3,306/QALY). For highest-risk patients, if empiric trimethoprimsalfamethoxazole was considered entirely to be outpatient
Received from the Sectionof General Internal Medicine and the Clinical AIDSProgram (KAF), Department of Medicine and Thorndike Memorial Laboratory, Boston City Hospital, Boston University School of Medicine; the Division of Clinical Epidemiology (ANAT, LG), Department of Medicine, Brigham and Women's Hospital and Beth Israel Hospital, Harvard Medical School; the Department of Health Policyand Management (A_NAT,DJC), HarvardSchoolof Public Health; and the Divisionof Infectious Disease (DJC), Department of Medicine, Beth Israel Hospital, Harvard Medical School, Boston, Massachusetts. Supported in part by a training grant for FacultyDevelopmentin General Internal Medicine (National Research Service Award 5T32DK07603-02) and a grant from the Agency for Health Care Policy and Research (1 -R0I-HS06694). Address correspondence and reprint requests to Dr. Freedberg: Section of General Internal Medicine, ACC-3,Boston City Hospital, 818 Harrison Avenue, Boston, MA02118.
therapy, it was preferred management i f the probabiltty o f PCP was above 38%. Conclusions: The authors conclude that preferred management strategies are determined more by differences in costs than by differences in levels o f effectiveness, and that they vary depending on the probability o f PCP in definable patient subgroups. Key words: AIDS; HIV,. Pneumocystis carinii pneumon/a; cost-effectiveness; decision analysis. J GEN INTERN MED
1992;7:261 - 272. THERE ARE AN ESTIMATED one million p e o p l e in the United States infected with the h u m a n immunodeficiency virus (HIV). x Over 2 1 0 , 0 0 0 have developed AIDS, and half of those have died. 2-3 Respiratory complications are among the most c o m m o n presenting manifestations of AIDS. P n e u m o c y s t i s c a r i n i i p n e u m o nia has accounted for approximately 70% of respiratory presentations, while the remaining 30% have been due to a variety of infections and neoplasms.4-6 Although the widespread use of zidovudine (AZT) and prophylaxis for Pneumocystis infection appears to be decreasing the n u m b e r of cases of P n e u m o c y s t i s c a r i n i i p n e u m o nia, it remains a c o m m o n cause of morbidity and mortality for individuals infected with HIV. TM The optimal m e t h o d of diagnosis and management o f patients at risk for Pneumocystis infection has not been clearly defined. Many diagnostic and prognostic tests are available, including chest radiography, arterial b l o o d gas analysis, p u l m o n a r y function testing, gallium scanning, i n d u c e d s p u t u m examination, and bronchosc o p y with bronchoalveolar lavage. Standard therapy includes oral or intravenous trimethoprim sulfamethoxazole and intravenous pentamidine, both of w h i c h are associated with significant rates o f adverse reactions in this patient population. 4,6,1s-2~ Newer therapies, including oral t r i m e t h o p r i m - d a p s o n e , are undergoing intense investigation, though none has been shown to be more effective than standard therapy. 22-26 The combination of effective therapy and difficult diagnosis presents a c o m m o n clinical dilemma: what is the appropriate management of HIV-infected patients w h o present with p u l m o n a r y symptoms? Several approaches have been suggested in the literature, but without validation. 4,27-29 We therefore developed a 261
Z62
Freedberg etal., COST-EFFECTIVEHIV MANAGEMENT
c o m p r e h e n s i v e decision-analytic m o d e l for patients at risk for AIDS-related Pneumocystis carinii pneumonia w h o present w i t h at least one w e e k of c o u g h or dyspnea. We e x a m i n e d differences in qualityadjusted life expectancy, direct medical costs, and c o s t - e f f e c t i v e n e s s for various clinical m a n a g e m e n t strategies.
METHODS Structure of the Model We d e v e l o p e d a decision tree m o d e l 3° to identify clinical strategies that m a x i m i z e quality-adjusted life e x p e c t a n c y in HIV-infected persons w h o present with c o u g h or dyspnea. We considered three types of patients based on their histories: 1) patients with k n o w n total CD4 ( h e l p e r l y m p h o c y t e ) counts b e l o w 2 0 0 / m m 3, 2) patients with k n o w n CD4 counts from 200 to 5 0 0 / m m 3, and 3) patients with total CD4 counts above 5 0 0 / m m 3. Each g r o u p was further subdivided according to initial chest radiograph findings. Abnormal chest radiographs w e r e defined as those with findings suggestive of an infectious process consistent w i t h Pneu. mocystis carinii p n e u m o n i a . Since most of the data in the literature are based on men, the prototypical patient was considered to be a 35-year-old man. The diagnostic tests considered w e r e arterial b l o o d gas analysis for abnormal arterial/alveolar (A/a) gradient ( > 15 m m Hg), induced s p u t u m examination with toluidine b l u e or giemsa stain for Pneumocystis, and b r o n c h o s c o p y with bronchoalveolar lavage. Although induced s p u t u m analysis and b r o n c h o s c o p y are used primarily for diagnosis, arterial b l o o d gas analysis may be used to assist b o t h diagnosis (since an abnormal A/a gradient increases the probability that a patient has Pneumocystis infection) and prognosis as a measure of physiologic dysfunction. Because arterial b l o o d gas analysis is less specific than the other tests for Pneumocystis carinii p n e u m o n i a , it offers less diagnostic information in the model. However, since the results of arterial b l o o d gas analysis are available m u c h m o r e quickly than are those of the o t h e r tests, w e have considered their impacts on m a n a g e m e n t separately, even though in m a n y situations o t h e r tests m a y be d o n e concurrently. We assumed that tests w e r e conditionally i n d e p e n d e n t 3~ and that tests c o u l d be done once in any sequence, w i t h the following limitation: no further testing was p e r f o r m e d following either a positive ind u c e d s p u t u m stained for Pneumocystis or a bronchoscopy. Treatments considered for Pneumocystis carinii p n e u m o n i a w e r e p e n t a m i d i n e (4 mg p e r kg o f b o d y weight, administered intravenously o n c e a day for 21 days) and t r i m e t h o p r i m - s u l f a m e t h o x a z o l e (20 mg trim e t h o p r i m and 100 m g sulfamethoxazole p e r kg of b o d y weight, administered intravenously in four divided doses a day, or two double-strength tablets ad-
ministered orally four times a day, for 21 days). Initial treatment for bronchitis or other p n e u m o n i a was erythromycin ( 5 0 0 m g orally four times a day for seven days). The analysis was designed to consider the evaluation and m a n a g e m e n t of patients whose conditions do not necessitate admission other than for treatment of p u l m o n a r y disease; w e did not consider patients with unstable vital signs for w h o m outpatient erythromycin w o u l d be inappropriate. Each test and t r e a t m e n t was assumed to require s o m e time and to cause some discomfort and, hence, to result in a d e c r e m e n t in qualityadjusted life expectancy, w h i c h was subtracted from the total life e x p e c t a n c y of any strategy of w h i c h i t w a s a part. All patients w e r e ultimately considered to fall into one of four diagnostic groups: Pneumocystis carinii p n e u m o n i a ; other AIDS-related diagnoses, including but not limited to cytomegalovirus, Mycobacterium avium intracellulare infection, and Kaposi's sarcoma; infections that are not necessarilyAIDS-related but that m a y be m o r e c o m m o n in AIDS patients, including bacterial infections and bronchitis; and noninfectious conditions. For the p u r p o s e s of this analysis, these groups w e r e considered to be all-inclusive and m u t u a l l y exclusive. Although Pneumocystis is c o m m o n l y seen in conjunction w i t h other infectious agents, w e considered it to be the primary p a t h o g e n w h e n found, so that life e x p e c t a n c y d e p e n d e d on its a p p r o p r i a t e treatment. Survival estimates w e r e based on patients' likelihoods of surviving the 21 -day p e r i o d after presentation, given their diagnoses and treatments. Patients w h o had survived this initial p e r i o d w e r e then assigned life expectancies based on w h i c h of the three initial HlVstatus groups they w e r e in, as well as on their current diagnoses. Estimates for the analysis w e r e obtained from the medical literature w h e n e v e r available. Baseline estimates and ranges not obtained f r o m p u b l i s h e d sources are those of one of the authors (DJC) (shown in Appendix A).
Cost- Effectiveness We defined 30 different clinical strategies made u p of the tests and treatments described above in various sequences and combinations. For each strategy w e used the m o d e l to project total direct medical costs (including the costs o f tests, medications, complications, hospital care, and h o m e care) and quality-adjusted life expectancies in a formal c o s t - e f f e c t i v e n e s s analysis. 3z Both average c o s t - e f f e c t i v e n e s s ratios, w h i c h c o m p a r e the costs and life e x p e c t a n c i e s of each strategy w i t h those of a strategy of " d o nothing," and incremental c o s t - e f f e c t i v e n e s s ratios, w h i c h measure the added costs and years of life gained for successively m o r e effective strategies in relation to one another, w e r e c o m p u t e d . We considered any strategy w i t h an
263
JOURNALOFGENERALINTERNALMEDICINE,Volume 7 (May/June), 1992
incremental c o s t - effectiveness ratio b e l o w $ 5 0 , 0 0 0 per quality-adjusted life year (QALY) saved to be potentially reasonable, and w e defined the r e c o m m e n d e d strategy as the one associated with the longest life expectancy among all reasonable strategies. 33 However, since the $ 5 0 , 0 0 0 threshold is somewhat arbitrary and may not be universally accepted, w e also examined the impact of using $20,000 and $100,000 per QALY saved as the threshold for the r e c o m m e n d e d strategy. Actual resource costs w e r e used in the analysis. Costs were determined from hospital charge data, w h i c h w e r e converted to costs by the use of costcenter-specific cost-to-charge ratios. The analysis was done from a societal perspective, w h i c h considered all direct medical costs, regardless of the payer. We did not include indirect costs in the analysis, because there are few data available regarding indirect costs related to these diagnoses in this population. All costs and life expectancies were discounted at a rate of 5% per year, thereby placing less value on years of life occurring in the distant future. 32 Costs utilized in the model are shown in Table 1.
Data and Assumptions
Diagnoses. The initial distribution of diagnoses for HIV-infected patients presenting with persistent cough or dyspnea was based on which of six groups the patient fell into, depending on CD4 c o u n t and initial chest radiograph results, and on three baseline assumptions. First, we assumed that without the use of zidovudine or Pneumocystis prophylaxis, P n e u m o c y s t i s cari n i i p n e u m o n i a w o u l d comprise 65%, 25%, and 2% of the causes of respiratory disease in patients with abnormal chest radiographs and < 200, 200 - 500, and > 500 CD4 c e l l s / m m 3, respectively. 4"s Second, we assumed that zidovudine w o u l d decrease the probability of
TABLE 1 Medical Costs in the Analysis* Intervention
Base ($)
Range ($)
Arterial blood gas analysis Induced sputum analysis Bronchoscopy with lavage TMP- SMXt, oral, 3 weeks TMP- SMX, intravenous, 1 day Pentamidine, intravenous, 1 day Erythromycin, oral, 1 week Daily hospital bed Daily home care, intravenous
17 43 743 38 58 129 3 370 92
5 - 1O0 5-150 100- 1,400 5 - 100 20 - 200 20-300 0-40 5 0 - 1,200 O- 300
*All costs from the Beth Israel Hospital, Boston, MA. 3s. ~ t T M P - SMX = trirnethoprim- sulfamethoxazole.
Pneumocystis infection and other AIDS diagnoses by 10%, and that Pneumocystis prophylaxis w o u l d decrease the probability of infection with Pneumocystis by 33%. 7it Third, we assumed that in patients with abnormal chest radiographs, 95% of the non-AIDS diagnoses w o u l d be infectious (e.g., bacterial pneumonia), while in patients with normal chest radiographs, 90% of the non-AIDS diagnoses w o u l d be infectious (e.g., bronchitis). The initial estimated distribution of diagnoses for each patient group is shown in Table 2. While these estimates were meant to be representative of different patient groups, these data are not available in the literature, and further information, such as clinical history and specific type of chest radiograph abnormality, w o u l d in many cases substantially change the likelihood that a given patient had Pneumocystis infection. We therefore varied all of these estimates in sensitivity analyses described below.
Diagnostic Tests. Baseline characteristics for arterial blood gas analysis included a sensitivity for Pneumocystis of 90% and a specificity of 29%; 4, 5, 35 for
TABLE 2 Estimated Initial Distribution of Diagnoses Based on HIV Status and Chest Radiograph
PCP*
Other AIDS Diagnosis
(%)
(%)
Infectious (%)
Noninfectious (%)
CD4t < 200/ram 3 or prior PCP* Chest radiograph abnormal Chest radiograph normal
39.2 24.1
38.9 22.6
20.8 48.0
1.1 5.3
CD4 2 0 0 - BOO/mm3§ Chest radiograph abnormal Chest radiograph normal
22.5 13.5
9.1 9.3
65.0 69.5
3.4 7.7
CD4 > 500/ram ~ Chest radiograph abnormal Chest radiograph normal
2.0 0.5
1.0 0.5
92.2 89.1
4.8 9.9
*PCP = Pneumocystis carinii pneumonia. tCD4 = total CD4-positive lymphocyte count. Distribution based on patient receiving zidovudine and some form of prophylaxis for Pneumocystis carinii pneumonia. §Distribution based on patient receiving zidovudine.
Z64
Freedberg et al., COST-EFFECTIVEHIV MANAGEMENT
induced sputum analysis, a sensitivity of 55% and a specificity of 99%; 36.39 and for b r o n c h o s c o p y with bronchoalveolar lavage, a sensitivity of 95% and a specificity of 99.5%.4, 29, 40.43 Quality-of-life decrements for arterial blood gas analysis, induced sputum analysis, and b r o n c h o s c o p y with lavage w e r e estimated at O. 1, 0.2, and 1.0 days, respectively.
efficacy of the first anti-Pneumocystis therapy used. Relative survival was decreased by 8% if therapy had been changed to a second anti-Pneumocystis drug because of complications, or by 40% if the first therapy had not been tolerated and no second therapy had b e e n instituted.4, 45 Relative survival was decreased by 5% (range O - 50%) among patients with Pneumocystis infection for w h o m appropriate therapy had been delayed for seven days because of empiric erythromycin.12, 44 The use of steroids as adjunctive therapy for Pneumocystis infection was not m o d e l e d directly; however, the baseline survival rate of 90% (range 70% to 98%) for t r i m e t h o p r i m - s u l f a m e t h o x a z o l e or intravenous pentamidine was consistent with recent data using steroids. 46, 47 Treatment for AIDS-related diagnoses other than infection with Pneumocystis was not specifically defined in the mode/; however, w e assumed that 10% of these diagnoses could be both detected by b r o n c h o s c o p y and treated effectively.4, 41, 43 Patients with Pneumocystis infection w h o had survived the initial 21-day period, and w h o had received effective therapy for Pneumocystis infection, were assigned a life e x p e c t a n c y of 657 days (range 328 to 1,314 days); those with other AIDS-related pulmonary diagnoses were assigned a life e x p e c t a n c y o f 648 days (range 324 to 1,296 days). For patients with the most severe immune suppression (CD4 < 2 0 0 / m m 3 ) , w e used the lower range o f survival in the baseline analysis.48, 49 Patients with CD4 2 0 0 - 5 0 0 / m m 3 w e r e assumed to have average life expectancies 2.00 years longer than those o f patients with CD4 < 2 0 0 / m m 3, and patients with CD4 > 5 0 0 / m m 3 were assumed to have life expectancies 5.00 years longer than those o f patients with CD4 < 200/mm3. 5°, sl Because there are no firm data in the literature on many of the model
Treatments. Complications from treatment inc l u d e d both progression of disease (nonresponse to therapy) and adverse drug reactions. Minor complications were assumed to necessitate only a change in therapy, while major complications were assumed to necessitate hospitalization for an average of four days. For patients with Pneumocystis infection, the baseline treatment characteristics and medication costs were estimated as follows: for t r i m e t h o p r i m sulfamethoxazole, a 21-day survival rate of 90% and a total complication rate necessitating a change in therapy o f 50%; for pentamidine, a 21-day survival rate o f 90% and a total complication rate of 40%; and for erythromycin, a 21-day survival of 66% and a complication rate o f 100%. 4, 6, 15, 16, 18, 20, 44 The baseline analysis considered therapy for Pneumocystis infection to be four days of inpatient intravenous therapy ( t r i m e t h o p r i m - s u l f a m e t h o x a z o l e or pentamidine) followed by 17 days of outpatient therapy (oral t r i m e t h o p r i m - s u l f a m e t h o x a z o l e or intravenous pentamidine through h o m e care). If empiric t r i m e t h o p r i m - s u l f a m e t h o x a z o l e or pentamidine was given before testing, we assumed that the relative sensitivities of all tests w o u l d be decreased by 10%.
Survival and Life Expectancy. Survival among patients with Pneumocystis infection was based on the
TABLE 3 Life Expectanciesin DiscountedQuality-adjustedLife Years: Optimal versus Worst First Step in Management(Not Including "Do Nothing") Quality-adJusted Life Expectancy Optimal First Step
Worst Years
First Step
Years
% Difference*
CD4 < 200/mm 3, or prior PCPt Chest radiograph abnormal Chest radiograph normal
Sputum* Sputum
1.10 1.55
TMP- SMX§ TMP- SMX
1.06 1.$2
3.8 2.0
CD4 200- 500/ram ~ Chest radiographabnormal Chest radiograph normal
Sputum Sputum
3.17 3.40
ABG¶ TMP-SMX
3.12 3.37
1,6 0,9
CD4 > 500/ram 3 Chest radiographabnormal Chest radiograph normal
Erythromycinll Erythromycin
6.10 6.22
ABG TMP-SMX
5.96 6.21
2,3 0,2
*% difference calculatedas [(optimal -- worst)/worst]. t PCP ----Pneumocystiscarin#pneumonia, *Sputum = strategy beginningwith inducedsputum analysis. §TMP-SMX = strategy beginningwith empiric trimethoprim-sulfamethoxazole. ¶ABG = strategy beginningwith arterial blood gas analysis,followed by intravenouspentamidine if abnormal. I}Erythromycin = strategy beginningwith empiric erythromydn,
JOURNALOFGENERALINTERNALMEDICINE,Volume 7 (May/June), 1992
265
TABLE 4 Total ExpectedCosts Basedon First Step in Management:Least ExpensiveComparedwith Most ExpensiveStrategies Expected Cost of Strategy Least Expensive
Most Expensive
First Step
Cost ($)
First Step
Cost ($)
CD4 < 200/mm 3, or prior PCPt Chest radiographabnormal Chest radiograph normal
Erythromycin$ Erythromycin
2,183 1,334
ABG§ ABG
4,065 2,977
123
CD4 200- 500/mm 3 Chest radiographabnormal Chest radiograph normal
Erythromycin Erythromycin
1,117 718
ABG TMP- SMX¶
3,640 2,553
226 256
CD4 > 500/mm 3 Chest radiographabnormal Chest radiograph normal
Erythromycin Erythromycin
114 42
ABG TMP- SMX
3,830 2,257
3,260 5,274
% Difference* 86
* % differencecalculatedas [(most- least)/least]. t PCP = PneumocystT"scarinii pneumonia. $Erythromycin = strategy beginningwith empiric erythromycin. §ABG = strategy beginningwith arterial blood gas analysis,followed by intravenouspentamidineif abnormal. ¶TMP-SMX = strategy beginningwith empiric trimethoprim- sulfamethoxazole.
parameters, extensive sensitivity analyses were done to identify w h i c h parameters had the greatest impact on life e x p e c t a n c y and c o s t - effectiveness ratios. Baseline estimates and ranges considered in sensitivity analyses for selected model parameters are shown in App e n d i x A.
strategies associated with optimal and worst life expectancies ranged from 3.8% in the sickest patients (CD4 < 2 0 0 / r a m 3, abnormal chest radiographs) to 0.2% in the least ill patients (CD4 > 5 0 0 / m m 3, normal chest radiographs). Total Costs
RESULTS Quality-adjusted Life Expectancy If no treatment was given to any patient, the estimated discounted life expectancies for the various patient groups ranged from 0.81 years (296 days) among CD4 < 2 0 0 / m m 3 patients with abnormal chest radiographs to 6.21 years among patients with CD4 > 5 0 0 / m m 3 and normal chest radiographs. For each patient type, the strategy that maximized life e x p e c t a n c y was identified. The most dramatic improvement attributable to optimal management was in CD4 < 2 0 0 / m m 3 patients with abnormal chest radiographs, whose average life expectancies increased from 296 days with the " d o nothing" strategy to 402 days with optimal management, an increase of 36%. Patients with abnormal chest radiographs benefitted more from optimal management than did patients at the same stage of illness with normal chest radiographs. The stepwise nature of the strategies in the decision model simulated clinical management by allowing the evaluation and treatment process to continue until appropriate therapy was found to optimize the likelihood o f a successful o u t c o m e for each patient. Because of this approach, there was very little difference in life expectancies w h e n strategies with alternative first steps in management were compared with each other (Table 3). The differences in life e x p e c t a n c y between
In contrast to life expectancy, the total costs of different strategies varied markedly. For example, for CD4 < 2 0 0 / m m 3 patients with abnormal chest radiographs, the average costs ranged from $ 2,183 for a strategy beginning with erythromycin to $4,065 for a strategy beginning with arterial b l o o d gas analysis followed by intravenous pentamidine if abnormal, an 86% increase in cost. Yet this arterial b l o o d gas strategy yielded a quality-adjusted life e x p e c t a n c y that was only 0.6 days (0.1%) longer than that of the strategy that began with erythromycin. For the other patient groups, the cost differences were even greater, ranging from 123% to 5,274% (Table 4). C o s t - Effectiveness
Among all patient groups, there were three strategies that were consistently most cost-effective and had incremental c o s t - effectiveness ratios b e l o w $ 5 0 , 0 0 0 per QALY saved. Table 5 indicates these strategies for each group of patients. For CD4 < 2 0 0 / m m 3 patients with a b n o r m a l chest radiographs (patients with the highest prior probability of Pneumocystis infection), the preferred strategy began with examination of an induced sputum specimen (Strategy 1, Fig. 1) and was associated with an average life e x p e c t a n c y of 402 days, average cost of
266
Freedberg etaL, COST-EFFECTIVEHIV MANAGEMENT
TABLE S Preferred Strategies and Other Strategies with Incremental Cost-Effectiveness Ratios
Received from the Sectionof General Internal Medicine and the Clinical AIDSProgram (KAF), Department of Medicine and Thorndike Memorial Laboratory, Boston City Hospital, Boston University School of Medicine; the Division of Clinical Epidemiology (ANAT, LG), Department of Medicine, Brigham and Women's Hospital and Beth Israel Hospital, Harvard Medical School; the Department of Health Policyand Management (A_NAT,DJC), HarvardSchoolof Public Health; and the Divisionof Infectious Disease (DJC), Department of Medicine, Beth Israel Hospital, Harvard Medical School, Boston, Massachusetts. Supported in part by a training grant for FacultyDevelopmentin General Internal Medicine (National Research Service Award 5T32DK07603-02) and a grant from the Agency for Health Care Policy and Research (1 -R0I-HS06694). Address correspondence and reprint requests to Dr. Freedberg: Section of General Internal Medicine, ACC-3,Boston City Hospital, 818 Harrison Avenue, Boston, MA02118.
therapy, it was preferred management i f the probabiltty o f PCP was above 38%. Conclusions: The authors conclude that preferred management strategies are determined more by differences in costs than by differences in levels o f effectiveness, and that they vary depending on the probability o f PCP in definable patient subgroups. Key words: AIDS; HIV,. Pneumocystis carinii pneumon/a; cost-effectiveness; decision analysis. J GEN INTERN MED
1992;7:261 - 272. THERE ARE AN ESTIMATED one million p e o p l e in the United States infected with the h u m a n immunodeficiency virus (HIV). x Over 2 1 0 , 0 0 0 have developed AIDS, and half of those have died. 2-3 Respiratory complications are among the most c o m m o n presenting manifestations of AIDS. P n e u m o c y s t i s c a r i n i i p n e u m o nia has accounted for approximately 70% of respiratory presentations, while the remaining 30% have been due to a variety of infections and neoplasms.4-6 Although the widespread use of zidovudine (AZT) and prophylaxis for Pneumocystis infection appears to be decreasing the n u m b e r of cases of P n e u m o c y s t i s c a r i n i i p n e u m o nia, it remains a c o m m o n cause of morbidity and mortality for individuals infected with HIV. TM The optimal m e t h o d of diagnosis and management o f patients at risk for Pneumocystis infection has not been clearly defined. Many diagnostic and prognostic tests are available, including chest radiography, arterial b l o o d gas analysis, p u l m o n a r y function testing, gallium scanning, i n d u c e d s p u t u m examination, and bronchosc o p y with bronchoalveolar lavage. Standard therapy includes oral or intravenous trimethoprim sulfamethoxazole and intravenous pentamidine, both of w h i c h are associated with significant rates o f adverse reactions in this patient population. 4,6,1s-2~ Newer therapies, including oral t r i m e t h o p r i m - d a p s o n e , are undergoing intense investigation, though none has been shown to be more effective than standard therapy. 22-26 The combination of effective therapy and difficult diagnosis presents a c o m m o n clinical dilemma: what is the appropriate management of HIV-infected patients w h o present with p u l m o n a r y symptoms? Several approaches have been suggested in the literature, but without validation. 4,27-29 We therefore developed a 261
Z62
Freedberg etal., COST-EFFECTIVEHIV MANAGEMENT
c o m p r e h e n s i v e decision-analytic m o d e l for patients at risk for AIDS-related Pneumocystis carinii pneumonia w h o present w i t h at least one w e e k of c o u g h or dyspnea. We e x a m i n e d differences in qualityadjusted life expectancy, direct medical costs, and c o s t - e f f e c t i v e n e s s for various clinical m a n a g e m e n t strategies.
METHODS Structure of the Model We d e v e l o p e d a decision tree m o d e l 3° to identify clinical strategies that m a x i m i z e quality-adjusted life e x p e c t a n c y in HIV-infected persons w h o present with c o u g h or dyspnea. We considered three types of patients based on their histories: 1) patients with k n o w n total CD4 ( h e l p e r l y m p h o c y t e ) counts b e l o w 2 0 0 / m m 3, 2) patients with k n o w n CD4 counts from 200 to 5 0 0 / m m 3, and 3) patients with total CD4 counts above 5 0 0 / m m 3. Each g r o u p was further subdivided according to initial chest radiograph findings. Abnormal chest radiographs w e r e defined as those with findings suggestive of an infectious process consistent w i t h Pneu. mocystis carinii p n e u m o n i a . Since most of the data in the literature are based on men, the prototypical patient was considered to be a 35-year-old man. The diagnostic tests considered w e r e arterial b l o o d gas analysis for abnormal arterial/alveolar (A/a) gradient ( > 15 m m Hg), induced s p u t u m examination with toluidine b l u e or giemsa stain for Pneumocystis, and b r o n c h o s c o p y with bronchoalveolar lavage. Although induced s p u t u m analysis and b r o n c h o s c o p y are used primarily for diagnosis, arterial b l o o d gas analysis may be used to assist b o t h diagnosis (since an abnormal A/a gradient increases the probability that a patient has Pneumocystis infection) and prognosis as a measure of physiologic dysfunction. Because arterial b l o o d gas analysis is less specific than the other tests for Pneumocystis carinii p n e u m o n i a , it offers less diagnostic information in the model. However, since the results of arterial b l o o d gas analysis are available m u c h m o r e quickly than are those of the o t h e r tests, w e have considered their impacts on m a n a g e m e n t separately, even though in m a n y situations o t h e r tests m a y be d o n e concurrently. We assumed that tests w e r e conditionally i n d e p e n d e n t 3~ and that tests c o u l d be done once in any sequence, w i t h the following limitation: no further testing was p e r f o r m e d following either a positive ind u c e d s p u t u m stained for Pneumocystis or a bronchoscopy. Treatments considered for Pneumocystis carinii p n e u m o n i a w e r e p e n t a m i d i n e (4 mg p e r kg o f b o d y weight, administered intravenously o n c e a day for 21 days) and t r i m e t h o p r i m - s u l f a m e t h o x a z o l e (20 mg trim e t h o p r i m and 100 m g sulfamethoxazole p e r kg of b o d y weight, administered intravenously in four divided doses a day, or two double-strength tablets ad-
ministered orally four times a day, for 21 days). Initial treatment for bronchitis or other p n e u m o n i a was erythromycin ( 5 0 0 m g orally four times a day for seven days). The analysis was designed to consider the evaluation and m a n a g e m e n t of patients whose conditions do not necessitate admission other than for treatment of p u l m o n a r y disease; w e did not consider patients with unstable vital signs for w h o m outpatient erythromycin w o u l d be inappropriate. Each test and t r e a t m e n t was assumed to require s o m e time and to cause some discomfort and, hence, to result in a d e c r e m e n t in qualityadjusted life expectancy, w h i c h was subtracted from the total life e x p e c t a n c y of any strategy of w h i c h i t w a s a part. All patients w e r e ultimately considered to fall into one of four diagnostic groups: Pneumocystis carinii p n e u m o n i a ; other AIDS-related diagnoses, including but not limited to cytomegalovirus, Mycobacterium avium intracellulare infection, and Kaposi's sarcoma; infections that are not necessarilyAIDS-related but that m a y be m o r e c o m m o n in AIDS patients, including bacterial infections and bronchitis; and noninfectious conditions. For the p u r p o s e s of this analysis, these groups w e r e considered to be all-inclusive and m u t u a l l y exclusive. Although Pneumocystis is c o m m o n l y seen in conjunction w i t h other infectious agents, w e considered it to be the primary p a t h o g e n w h e n found, so that life e x p e c t a n c y d e p e n d e d on its a p p r o p r i a t e treatment. Survival estimates w e r e based on patients' likelihoods of surviving the 21 -day p e r i o d after presentation, given their diagnoses and treatments. Patients w h o had survived this initial p e r i o d w e r e then assigned life expectancies based on w h i c h of the three initial HlVstatus groups they w e r e in, as well as on their current diagnoses. Estimates for the analysis w e r e obtained from the medical literature w h e n e v e r available. Baseline estimates and ranges not obtained f r o m p u b l i s h e d sources are those of one of the authors (DJC) (shown in Appendix A).
Cost- Effectiveness We defined 30 different clinical strategies made u p of the tests and treatments described above in various sequences and combinations. For each strategy w e used the m o d e l to project total direct medical costs (including the costs o f tests, medications, complications, hospital care, and h o m e care) and quality-adjusted life expectancies in a formal c o s t - e f f e c t i v e n e s s analysis. 3z Both average c o s t - e f f e c t i v e n e s s ratios, w h i c h c o m p a r e the costs and life e x p e c t a n c i e s of each strategy w i t h those of a strategy of " d o nothing," and incremental c o s t - e f f e c t i v e n e s s ratios, w h i c h measure the added costs and years of life gained for successively m o r e effective strategies in relation to one another, w e r e c o m p u t e d . We considered any strategy w i t h an
263
JOURNALOFGENERALINTERNALMEDICINE,Volume 7 (May/June), 1992
incremental c o s t - effectiveness ratio b e l o w $ 5 0 , 0 0 0 per quality-adjusted life year (QALY) saved to be potentially reasonable, and w e defined the r e c o m m e n d e d strategy as the one associated with the longest life expectancy among all reasonable strategies. 33 However, since the $ 5 0 , 0 0 0 threshold is somewhat arbitrary and may not be universally accepted, w e also examined the impact of using $20,000 and $100,000 per QALY saved as the threshold for the r e c o m m e n d e d strategy. Actual resource costs w e r e used in the analysis. Costs were determined from hospital charge data, w h i c h w e r e converted to costs by the use of costcenter-specific cost-to-charge ratios. The analysis was done from a societal perspective, w h i c h considered all direct medical costs, regardless of the payer. We did not include indirect costs in the analysis, because there are few data available regarding indirect costs related to these diagnoses in this population. All costs and life expectancies were discounted at a rate of 5% per year, thereby placing less value on years of life occurring in the distant future. 32 Costs utilized in the model are shown in Table 1.
Data and Assumptions
Diagnoses. The initial distribution of diagnoses for HIV-infected patients presenting with persistent cough or dyspnea was based on which of six groups the patient fell into, depending on CD4 c o u n t and initial chest radiograph results, and on three baseline assumptions. First, we assumed that without the use of zidovudine or Pneumocystis prophylaxis, P n e u m o c y s t i s cari n i i p n e u m o n i a w o u l d comprise 65%, 25%, and 2% of the causes of respiratory disease in patients with abnormal chest radiographs and < 200, 200 - 500, and > 500 CD4 c e l l s / m m 3, respectively. 4"s Second, we assumed that zidovudine w o u l d decrease the probability of
TABLE 1 Medical Costs in the Analysis* Intervention
Base ($)
Range ($)
Arterial blood gas analysis Induced sputum analysis Bronchoscopy with lavage TMP- SMXt, oral, 3 weeks TMP- SMX, intravenous, 1 day Pentamidine, intravenous, 1 day Erythromycin, oral, 1 week Daily hospital bed Daily home care, intravenous
17 43 743 38 58 129 3 370 92
5 - 1O0 5-150 100- 1,400 5 - 100 20 - 200 20-300 0-40 5 0 - 1,200 O- 300
*All costs from the Beth Israel Hospital, Boston, MA. 3s. ~ t T M P - SMX = trirnethoprim- sulfamethoxazole.
Pneumocystis infection and other AIDS diagnoses by 10%, and that Pneumocystis prophylaxis w o u l d decrease the probability of infection with Pneumocystis by 33%. 7it Third, we assumed that in patients with abnormal chest radiographs, 95% of the non-AIDS diagnoses w o u l d be infectious (e.g., bacterial pneumonia), while in patients with normal chest radiographs, 90% of the non-AIDS diagnoses w o u l d be infectious (e.g., bronchitis). The initial estimated distribution of diagnoses for each patient group is shown in Table 2. While these estimates were meant to be representative of different patient groups, these data are not available in the literature, and further information, such as clinical history and specific type of chest radiograph abnormality, w o u l d in many cases substantially change the likelihood that a given patient had Pneumocystis infection. We therefore varied all of these estimates in sensitivity analyses described below.
Diagnostic Tests. Baseline characteristics for arterial blood gas analysis included a sensitivity for Pneumocystis of 90% and a specificity of 29%; 4, 5, 35 for
TABLE 2 Estimated Initial Distribution of Diagnoses Based on HIV Status and Chest Radiograph
PCP*
Other AIDS Diagnosis
(%)
(%)
Infectious (%)
Noninfectious (%)
CD4t < 200/ram 3 or prior PCP* Chest radiograph abnormal Chest radiograph normal
39.2 24.1
38.9 22.6
20.8 48.0
1.1 5.3
CD4 2 0 0 - BOO/mm3§ Chest radiograph abnormal Chest radiograph normal
22.5 13.5
9.1 9.3
65.0 69.5
3.4 7.7
CD4 > 500/ram ~ Chest radiograph abnormal Chest radiograph normal
2.0 0.5
1.0 0.5
92.2 89.1
4.8 9.9
*PCP = Pneumocystis carinii pneumonia. tCD4 = total CD4-positive lymphocyte count. Distribution based on patient receiving zidovudine and some form of prophylaxis for Pneumocystis carinii pneumonia. §Distribution based on patient receiving zidovudine.
Z64
Freedberg et al., COST-EFFECTIVEHIV MANAGEMENT
induced sputum analysis, a sensitivity of 55% and a specificity of 99%; 36.39 and for b r o n c h o s c o p y with bronchoalveolar lavage, a sensitivity of 95% and a specificity of 99.5%.4, 29, 40.43 Quality-of-life decrements for arterial blood gas analysis, induced sputum analysis, and b r o n c h o s c o p y with lavage w e r e estimated at O. 1, 0.2, and 1.0 days, respectively.
efficacy of the first anti-Pneumocystis therapy used. Relative survival was decreased by 8% if therapy had been changed to a second anti-Pneumocystis drug because of complications, or by 40% if the first therapy had not been tolerated and no second therapy had b e e n instituted.4, 45 Relative survival was decreased by 5% (range O - 50%) among patients with Pneumocystis infection for w h o m appropriate therapy had been delayed for seven days because of empiric erythromycin.12, 44 The use of steroids as adjunctive therapy for Pneumocystis infection was not m o d e l e d directly; however, the baseline survival rate of 90% (range 70% to 98%) for t r i m e t h o p r i m - s u l f a m e t h o x a z o l e or intravenous pentamidine was consistent with recent data using steroids. 46, 47 Treatment for AIDS-related diagnoses other than infection with Pneumocystis was not specifically defined in the mode/; however, w e assumed that 10% of these diagnoses could be both detected by b r o n c h o s c o p y and treated effectively.4, 41, 43 Patients with Pneumocystis infection w h o had survived the initial 21-day period, and w h o had received effective therapy for Pneumocystis infection, were assigned a life e x p e c t a n c y of 657 days (range 328 to 1,314 days); those with other AIDS-related pulmonary diagnoses were assigned a life e x p e c t a n c y o f 648 days (range 324 to 1,296 days). For patients with the most severe immune suppression (CD4 < 2 0 0 / m m 3 ) , w e used the lower range o f survival in the baseline analysis.48, 49 Patients with CD4 2 0 0 - 5 0 0 / m m 3 w e r e assumed to have average life expectancies 2.00 years longer than those o f patients with CD4 < 2 0 0 / m m 3, and patients with CD4 > 5 0 0 / m m 3 were assumed to have life expectancies 5.00 years longer than those o f patients with CD4 < 200/mm3. 5°, sl Because there are no firm data in the literature on many of the model
Treatments. Complications from treatment inc l u d e d both progression of disease (nonresponse to therapy) and adverse drug reactions. Minor complications were assumed to necessitate only a change in therapy, while major complications were assumed to necessitate hospitalization for an average of four days. For patients with Pneumocystis infection, the baseline treatment characteristics and medication costs were estimated as follows: for t r i m e t h o p r i m sulfamethoxazole, a 21-day survival rate of 90% and a total complication rate necessitating a change in therapy o f 50%; for pentamidine, a 21-day survival rate o f 90% and a total complication rate of 40%; and for erythromycin, a 21-day survival of 66% and a complication rate o f 100%. 4, 6, 15, 16, 18, 20, 44 The baseline analysis considered therapy for Pneumocystis infection to be four days of inpatient intravenous therapy ( t r i m e t h o p r i m - s u l f a m e t h o x a z o l e or pentamidine) followed by 17 days of outpatient therapy (oral t r i m e t h o p r i m - s u l f a m e t h o x a z o l e or intravenous pentamidine through h o m e care). If empiric t r i m e t h o p r i m - s u l f a m e t h o x a z o l e or pentamidine was given before testing, we assumed that the relative sensitivities of all tests w o u l d be decreased by 10%.
Survival and Life Expectancy. Survival among patients with Pneumocystis infection was based on the
TABLE 3 Life Expectanciesin DiscountedQuality-adjustedLife Years: Optimal versus Worst First Step in Management(Not Including "Do Nothing") Quality-adJusted Life Expectancy Optimal First Step
Worst Years
First Step
Years
% Difference*
CD4 < 200/mm 3, or prior PCPt Chest radiograph abnormal Chest radiograph normal
Sputum* Sputum
1.10 1.55
TMP- SMX§ TMP- SMX
1.06 1.$2
3.8 2.0
CD4 200- 500/ram ~ Chest radiographabnormal Chest radiograph normal
Sputum Sputum
3.17 3.40
ABG¶ TMP-SMX
3.12 3.37
1,6 0,9
CD4 > 500/ram 3 Chest radiographabnormal Chest radiograph normal
Erythromycinll Erythromycin
6.10 6.22
ABG TMP-SMX
5.96 6.21
2,3 0,2
*% difference calculatedas [(optimal -- worst)/worst]. t PCP ----Pneumocystiscarin#pneumonia, *Sputum = strategy beginningwith inducedsputum analysis. §TMP-SMX = strategy beginningwith empiric trimethoprim-sulfamethoxazole. ¶ABG = strategy beginningwith arterial blood gas analysis,followed by intravenouspentamidine if abnormal. I}Erythromycin = strategy beginningwith empiric erythromydn,
JOURNALOFGENERALINTERNALMEDICINE,Volume 7 (May/June), 1992
265
TABLE 4 Total ExpectedCosts Basedon First Step in Management:Least ExpensiveComparedwith Most ExpensiveStrategies Expected Cost of Strategy Least Expensive
Most Expensive
First Step
Cost ($)
First Step
Cost ($)
CD4 < 200/mm 3, or prior PCPt Chest radiographabnormal Chest radiograph normal
Erythromycin$ Erythromycin
2,183 1,334
ABG§ ABG
4,065 2,977
123
CD4 200- 500/mm 3 Chest radiographabnormal Chest radiograph normal
Erythromycin Erythromycin
1,117 718
ABG TMP- SMX¶
3,640 2,553
226 256
CD4 > 500/mm 3 Chest radiographabnormal Chest radiograph normal
Erythromycin Erythromycin
114 42
ABG TMP- SMX
3,830 2,257
3,260 5,274
% Difference* 86
* % differencecalculatedas [(most- least)/least]. t PCP = PneumocystT"scarinii pneumonia. $Erythromycin = strategy beginningwith empiric erythromycin. §ABG = strategy beginningwith arterial blood gas analysis,followed by intravenouspentamidineif abnormal. ¶TMP-SMX = strategy beginningwith empiric trimethoprim- sulfamethoxazole.
parameters, extensive sensitivity analyses were done to identify w h i c h parameters had the greatest impact on life e x p e c t a n c y and c o s t - effectiveness ratios. Baseline estimates and ranges considered in sensitivity analyses for selected model parameters are shown in App e n d i x A.
strategies associated with optimal and worst life expectancies ranged from 3.8% in the sickest patients (CD4 < 2 0 0 / r a m 3, abnormal chest radiographs) to 0.2% in the least ill patients (CD4 > 5 0 0 / m m 3, normal chest radiographs). Total Costs
RESULTS Quality-adjusted Life Expectancy If no treatment was given to any patient, the estimated discounted life expectancies for the various patient groups ranged from 0.81 years (296 days) among CD4 < 2 0 0 / m m 3 patients with abnormal chest radiographs to 6.21 years among patients with CD4 > 5 0 0 / m m 3 and normal chest radiographs. For each patient type, the strategy that maximized life e x p e c t a n c y was identified. The most dramatic improvement attributable to optimal management was in CD4 < 2 0 0 / m m 3 patients with abnormal chest radiographs, whose average life expectancies increased from 296 days with the " d o nothing" strategy to 402 days with optimal management, an increase of 36%. Patients with abnormal chest radiographs benefitted more from optimal management than did patients at the same stage of illness with normal chest radiographs. The stepwise nature of the strategies in the decision model simulated clinical management by allowing the evaluation and treatment process to continue until appropriate therapy was found to optimize the likelihood o f a successful o u t c o m e for each patient. Because of this approach, there was very little difference in life expectancies w h e n strategies with alternative first steps in management were compared with each other (Table 3). The differences in life e x p e c t a n c y between
In contrast to life expectancy, the total costs of different strategies varied markedly. For example, for CD4 < 2 0 0 / m m 3 patients with abnormal chest radiographs, the average costs ranged from $ 2,183 for a strategy beginning with erythromycin to $4,065 for a strategy beginning with arterial b l o o d gas analysis followed by intravenous pentamidine if abnormal, an 86% increase in cost. Yet this arterial b l o o d gas strategy yielded a quality-adjusted life e x p e c t a n c y that was only 0.6 days (0.1%) longer than that of the strategy that began with erythromycin. For the other patient groups, the cost differences were even greater, ranging from 123% to 5,274% (Table 4). C o s t - Effectiveness
Among all patient groups, there were three strategies that were consistently most cost-effective and had incremental c o s t - effectiveness ratios b e l o w $ 5 0 , 0 0 0 per QALY saved. Table 5 indicates these strategies for each group of patients. For CD4 < 2 0 0 / m m 3 patients with a b n o r m a l chest radiographs (patients with the highest prior probability of Pneumocystis infection), the preferred strategy began with examination of an induced sputum specimen (Strategy 1, Fig. 1) and was associated with an average life e x p e c t a n c y of 402 days, average cost of
266
Freedberg etaL, COST-EFFECTIVEHIV MANAGEMENT
TABLE S Preferred Strategies and Other Strategies with Incremental Cost-Effectiveness Ratios
