Combined APACHE II Score and Serum Lactate Dehydrogenase as Predictors of In-hospital Mortality Caused by First Episode Pneumocystis carinii Pneumonia in Patients with Acquired Immunodeficiency Syndrome 1 , 2
CONSTANCE A. BENSON, JOEL SPEAR, DAVID HINES, JOHN C. POTTAGE, JR., HAROLD A. KESSLER, and GORDON M. TRENHOLME
Pneumocystis carinii pneumonia (PCP) has been the most commonly reported serious opportunistic infection in adult patients with acquired immunodeficiency syndrome (AIDS). It is the index diagnosis for 66070 of patients and occurs during the course of their illness in 80 to 85070 of patients with AIDS (1). Reported mortality rates associated with PCP range from 10 to 60070 (2-12). With the development of potentially more efficacious therapies for the treatment of patients with acute PCP, methods for predicting the outcome of a specific episode and for assuring comparability of disease severity will be important for stratification of patients in the design of clinical trials and for decision making regarding specific interventions such as salvage therapy. Several clinical and laboratory parameters have been evaluated as possible predictors of mortality and indirectly as measures of disease severity caused by PCP. These include increased alveolar-arterial oxygen gradient, low arterial oxygen and carbon dioxide tensions, leukocytosis, chest roentgenogram abnormalities, decreased serum albumin, increased serum lactate dehydrogenase (LDH), and concurrent pulmonary infections (3-14). The acute physiology and chronic health evaluation (APACHE II) scale is a physiologic severity of illness measurement devised to evaluate critically ill patients (15). An elevated APACHE II score has been shown to predict mortality for several acute disease states (15-17). Its utility in predicting mortality for patients with AIDS and PCP has not been established, and in at least one study, Smith and coworkers suggested the APACHE II system significantly underestimated
SUMMARY We retrospectively analyzed data from 75 hospitalized patients with a first episode of Pneumocyst/s catlnll pneumonia to compare the ability of four parameters, Including admission serum albumin, serum lactate dehydrogenase (LDH), alveolar-arterial oxygen gradient, and the APACHE II score, to predict mortality and response to Initial antlpneumocystls therapy. The eight patients who died due to pneumocystosls and the 12 who failed Initial antlpneumocystls therapy had significantly higher admission mean APACHEII scores and serum LDH levels and lower mean serum albumin levels than did the 65 who survived and the 61 who responded to Initial therapy (p < 0.05for each). Differences In mean alveolar..rterlal oxygen gradients were not statistically significant with respect to survival or response to Initial therapy. In a stepwise discriminant analysis of parameters associated with mortality, APACHE II score and LDH level were statistically significant (p < 0.0001 for each). In a stepwise discriminant analysis of parameters associated with response to Initial therapy, APACHE II score and LDH level"were again statistically significant (p < 0.0001, respectively). The addition of the alveolar-arterial oxygen gradient and serum albumin level did not further Increase the predictive ability of the discriminant analyses. When analyzed alone, neither the alveolar..rterlal oxygen gradient nor the serum albumin were statistically significant In each discriminant analysis. The APACHE II score combined with the serum LDH may be more useful than other parameters, singly or combined, to more closely match patients with regard to severity of IIIne88 due to first episode Pneumocyst/s catlnll pneumonia when comparing experimental new therapies with standard agents. AM REV RESPIR DIS 1991; 144:319-323
mortality when applied to patients with Methods AIDS and PCP who required mechaniPatient Data cal ventilation and intensive care unit The charts of all patients with AIDS hospitalmonitoring (13). An elevated serum LDH ized in Rush-Presbyterian-St. Luke's Medilevel has previously been suggested as an cal Center, Chicago, Illinois, with a histologindicator of more severe disease in these ically confirmed diagnosis of first episode patients (7-12). Although some studies Pneumocystiscarinii pneumonia (PCP) behave shown lower mean serum LDH lev- tween April 1, 1985 and December 31, 1987, els in survivors than in those who died, wereretrospectively reviewed. Data collected these have fallen short of establishing included risk factors for human immunodeficiency virus (HIV) infection, signs and LDH as an independent predictor of mortality (7-12). We retrospectively reviewed our experience treating persons (Received in original form May 30, 1990 and in with AIDS and a first episode of PCP revised form March 18, 1991) in order to establish a combination of 1 From the Departments of Immunology/Microlaboratory and clinical parameters, in- biology, Pharmacology, and Medicine, Section of cluding the APACHE II score combined Infectious Disease, Rush Medical Collegeand Rushwith the serum LDH level, which may Presbyterian-St. Luke's Medical Center, Chicago, be used to predict response to initial ther- Illinois. 2 Correspondence and requests for reprints apy, to predict mortality, and potentially should be addressed to Constance A. Benson, to assure comparability of disease severi- M.D., Section of Infectious Disease, Rush-Presty in the design and evaluation of clini- , byterian-St. Luke's Medical Center, 1653 WestCongress Parkway, Chicago, IL 60612. cal trials. 319
BENSON, SPEAR, HINES, POTTAGE, KESSLER, AND TRENHOLME
symptoms related to PCP, baseline arterial blood gas values while inspiring room air, percent inspired oxygen supplementation, complete blood counts, serum LDH, albumin and electrolytes, antipneumocystis therapy, response to initial antipneumocystis therapy, and survival or death. An APACHE II score was calculated based on data measured within 12 h of hospital admission for PCP as described by Knaus and coworkers (15). Briefly, patients were assigned a point value ranging from zero to four for abnormalities of each of 12 physiologic variables including mean arterial pressure, temperature, heart rate, respiratory rate, arterial oxygen concentration or alveolar-arterial oxygen gradient, arterial pH, serum sodium, potassium, creatinine, hematocrit, white blood cell count, and Glascow Coma Scale. Initial temperature, pulse, and respiratory rates measured at the time of presentation were used in the calculation. When mean arterial pressure could not be directly measured, a calculated value using the formula [diastolic cuff pressure - 1/3 (systolic cuff pressure - diastolic cuff pressure)] was substituted. The Glascow Coma Score was calculated using the formula [15 the actual Glascow Coma Score] (15).Because no patient had altered mental status at the time of admission, this value was zero for all included patients. In addition, points were assigned for age and the presence of underlying or chronic diseases. All patients received five points for their underlying immunosuppressive illness (AIDS) as defined in the method for calculating the APACHE II score (15). Response to initial therapy was defined as improvement in or resolution of symptoms associated with PCP without a change in initial therapy followed by discharge from the hospital without a requirement for ventilatory support at the time of discharge. Failure to respond to initial therapy was defined as clinical deterioration after at least 72 h oftherapy followed by change to an alternative antipneumocystis medication or death on initial therapy.
Statistical Analysis Data were analyzed using SPSS/PC +~ (SPSS, Inc., Chicago, IL). Baseline APACHE II score, serum LDH, serum albumin, and alveolar-arterial oxygen gradient on admission were used in stepwise discriminant analyses to obtain a set of variables for predicting survival and a set of variables for predicting response to initial therapy. Two-tailed pooledvariance and separate-variance t tests were used to compare patients who died with patients who survived with respect to APACHE II, serum LDH, serum albumin, and alveolararterial oxygen gradient on admission and were also used to compare patients who failed initial therapy with those who responded to initial therapy with respect to the same variables. One-way analysis of variance with Tukey confidence intervals was used to compare patients who responded to initial therapy and those who failed initial therapy with respect to the same mean values.
TABLE 1 RESPONSE TO INITIAL THERAPY WITH TRIMETHOPRIM-SULFAMETHOXAZOLE VERSUS PENTAMIDINE IN PATIENTS WITH AIDS AND FIRST EPISODE PCP Patient Response n
10 (7}t 2 (1)t
• Switched to an alternative therapy to which patients responded. t Number of deaths in parentheses.
Results initial therapy, the mean duration of Seventy-four men and one woman, with treatment prior to switching to an altera mean age of 36.7 yr (range 26 to 62) native agent was 7.6 ± 4.4 days (range were hospitalized with HIV infection and 3 to 17 days). There were two additional first episode PCP. Risk factors for HIV deaths, one caused by disseminated crypinfection werehomosexual/bisexual life- tococcal disease and the other by central style (69), blood product transfusion (2), nervous system lymphoma with cerebellar and heterosexual contact with a partner herniation, after both had responded to in a high risk group (2); two patients had PCP therapy. These two patients were exno determined risk factor. The mean hos- cluded from further analysis of mortality. The mean APACHE II score, mean sepital stay was 20.9 days (range 4 to 75 days). There were eight deaths (10070) at- rum LDH, mean alveolar-arterial oxytributed to PCP. Treatment regimens are gen gradient, and mean serum albumin summarized in table 1. Twenty-five pa- at the time of admission for those who tients were treated with trimethoprim- died because of PCP and those who sursulfamethoxazole alone, seven with pen- vived the episode appear in table 2. tamidine alone, and 43 with both agents Alveolar-arterial oxygen gradients were sequentially. The reasons for change of calculated during inspiration of room air therapy included toxicity(31 of 43, 72.1070) for all but two patients, one who survived or clinical failure of the initially admin- and one who died. These two were existered agent (12 of 43, 27.9070). A total cluded from the discriminant analysis. of 65 patients received trimethoprim- One additional patient was excluded sulfamethoxazole as initial therapy. For- from the discriminant analysis because ty of these 65 (61.5%) were switched to of missing data points for serum LDH an alternative therapy; 10 for failure of and albumin values. Those who died had initial therapy and 30 because of toxici- significantly higher mean APACHE II ty. Seven of the eight deaths occurred in scores and serum LDH values, and sigthe group initially treated with trimetho- nificantly lower mean serum albumin prim-sulfamethoxazole, Ten patients re- values. There was no statistically significeived pentamidine as initial therapy. cant difference between the mean value Three of these were changed to an alter- for alveolar-arterial oxygen gradient for native agent; two for failure and one for ' those who survived and those who died. In a stepwise discriminant analysis to toxicity. One death occurred among this group of 10patients. For those who failed predict mortality, APACHE II score and
TABLE 2 MEAN APACHE /I SCORE, SERUM LDH, SERUM ALBUMIN, AND ALVEOLAR-ARTERIAL OXYGEN GRADIENT FOR PATIENTS WHO SURVIVED VERSUS THOSE WHO DIED DUE TO FIRST EPISODE PCP· Survived APACHE II Serum LDH, U/L Serum albumin, g/dl Alveolar-arterial oxygen gradient, mm Hg
9.3 (n 350 (n 3.3 (n 43.4 (n
= 65) ± 144
= 65) ± 0.6
= 64) ± 38.6
• Values are mean :i: SO. Two-tailed pooled-variance and separate-variance t tests.
Died 14.8 (n 796 (n 2.7 (n 45 (n
± 2.0 8) ± 430 7) ± 0.7 8) ± 18.8 7)
< 0.004 0.033
APACHE II SCORE AND SERUM LDH AS MORTALITY PREDICTORS
TABLE 3 MEAN APACHE II SCORE, SERUM LDH, SERUM ALBUMIN AND ALVEOLAR-ARTERIAL OXYGEN GRADIENT FOR PATIENTS WHO FAILED INITIAL THERAPY VERSUS THOSE WHO RESPONDED OR DEVELOPED TOXICITY TO INITIAL THERAPY·
Parameter APACHE II LDH Albumin Alveolar-arterial oxygen gradient
Failed (n '" 12)
Respondedt (n '" 61)
12.3 ± 3.5 668 ± 365 2.6 ± 0.5
8.9 ± 4.0 339 ± 138 3.4 ± 0.1
0.004 0.002 < 0.001
49.6 ± 18.6
43.0 ± 15.5
• Values are mean ± so. t Includes 30 responders plus 31 developing toxicity responded. :j: Two-tailed pooled-variance and separate-variance
to initial therapy requiring a switch to an alternative therapy to which they
LDH were statistically significant (p < 0.0001 and p < 0.0001, respectively). Alveolar-arterial oxygen gradients and serum albumin were not statistically significant after APACHE II and LDH were included in the discriminant analysis. Using the discriminant function with APACHE II score and LDH, 90.28070 of the patients could be correctly classified with respect to mortality. Using the discriminant function with LDH and alveolar-arterial oxygen gradient, 89.86% of the patients could be correctly classified with respect to mortality. No single parameter analyzed independently using the discriminant function resulted in a higher percentage of correctly classified patients than the APACHE II score combined with LDH. Twelve patients failed initial therapy, 31 developed toxicity requiring a change in therapy, and 30 responded to initial therapy and required no changes. The mean values for APACHE II score, serum LDH, serum albumin, and alveolararterial oxygengradient for these patients are shown in table 3. Using two-tailed pooled-variance and separate-variance t tests, those who failed initial therapy had significantly higher mean APACHE II score and mean serum LDH and significantly lower mean serum albumin at the time of admission than those who responded to initial therapy or responded after a switch to alternative therapy. The difference between these two groups with respect to alveolar-arterial oxygen gradient was not statistically significant (p = 0.089). Using a stepwise discriminant analysis to predict response to initial therapy, APACHE II score and serum LDH were again statistically significant (p < 0.0001, respectively). Serum albumin and alveolar-arterial oxygen gradient were not statistically significant after APACHE II and LDH were included in the discriminant function; however, the discriminant
function was not as successful in correctly classifying patients according to therapeutic response as with mortality. Using the discriminant function with APACHE II and LDH, grouping patients according to failure of initial therapy or response to initial therapy, 68.29% were correctly classified. Discussion
The clinical presentation of PCP in patients with AIDS may range from mild derangement of oxygenation to diffuse alveolar damage with progressive respiratory failure (1-14). Such variability in severity of illness may impede the ability to predict outcome and response to treatment and to compare different therapeutic strategies. One method of evaluating severity of illness is the APACHE II scoring system (15). For a variety of disease states, including respiratory failure secondary to pneumonia, a consistent positive correlation has been validated for the APACHE II score and the risk of death (16, 17). In our retrospective study, the APACHE II score was statistically significant as one of two variables in a stepwise discriminant analysis for predicting survival with first episode PCP. In addition, scores were significantly higher for those who died or failed initial therapy than for those who survived or responded to initial therapy. These data suggest that failure of initial therapy and, ultimately, mortality may be related to severity of illness; however, we could not statistically validate the APACHE II score as a singleindependent predictor of those outcomes. Similarly, Smith and colleagues found the APACHE II score was not a statistically significant single predictor of mortality for 37 patients with AIDS, initial or recurrent PCP, and respiratory failure requiring mechanical ventilation (13). Their explanation was based on the lack of a specific diagnostic category coefficient that reflected PCP disease severity
in this population. Their study differed from ours in that weincluded 72 patients with only first episode PCP, and only 12 of our patients were mechanically ventilated requiring intensive care unit admission. Seven of the 12mechanically ventilated patients died. For this subgroup, the APACHE II score and serum LDH did not discriminate nonsurvivors from survivors(mean APACHE II, 12.4 and mean LDH, 659 for seven deaths versus mean APACHE II, 15.2 and mean LDH, 501 for five survivors; p > 0.10 for both). Similar to the APACHE II score, our data further demonstrate the statistical significance of serum LDH as one of two variables in a stepwise discriminant analysis predicting survival and demonstrate significantly higher LDH levelsfor those who died or who failed initial therapy. Serum LDH has been previously evaluated as a marker of disease severity in AIDS patients with PCP (6, 7). Kagawa and coworkers (7) suggested elevated serum LDH levelscorrelated with increased alveolar-arterial oxygen gradient; both correlated with PCP disease severity. They did not comment on LDH as a prognostic indicator or predictor of survival. Brenner and colleagues (5) reported that, when measured serially in patients with AIDS being treated for PCP, LDH values progressively declined in those who responded to therapy and progressivelyrose in those who failed. Efferen and coworkers (12) and El-Sadr and Simberkoff (14) found for those patients who required mechanical ventilation, admission and peak serum LDH did not discriminate between survivors and nonsurvivors. Zaman and White (11) noted an association between lowerserum LDH levelsand survival; however,they reported marked overlap of values precluding prediction of mortality for individual patients. Similar to our results, these data fall short of establishing an elevated serum LDH level as a single independent predictor of mortality. The alveolar-arterial oxygen gradient has been suggested as a useful measure of disease severity and an independent prognostic indicator of survival for persons with AIDS and acute PCP by a number of investigators (5, 9). Brenner and coworkers (5) found that the alveolar-arterial oxygen gradient correlated not only with probability of survival for the acute episode but that a low initial gradient strongly predicted longer term survival. Kales and colleagues (4), however, found 21 of 109 patients with normal alveolar-arterial oxygen gradients who did not show improved survival over
those with abnormal values. When evaluated in a stepwise discriminant analysis, alveolar-arterial oxygen gradient and albumin were not significant. Using twotailed pooled-variance and separate-variance t tests, alveolar-arterial oxygen gradients were not significantly higher for nonsurvivors than for survivors. Possible explanations for the disparity between previous reports and our data include the following. First, other studies included patients with one or more episodes of PCP (4, 5). Our study included only those with first episode PCP. Mortality rates for second or subsequent episodes of PCP havebeen reported to be higher than for first episode PCP (5). Second, Brenner and coworkers (5) did not obtain arterial blood gases consistently at the time of hospital admission as we did. Third, the former studies report data from patients treated between 1982 and 1986(4, 5). Mortality rates ranged from 19to 22070 in two studies evaluating first episode PCP to 30070 in the study by Brenner and colleagues, in which patients with multiple PCP episodes were included (4,5,9). Wharton and coworkers (2) reported variable mortality rates associated with first episode PCP, with quarterly rates ranging from 11 to 31% at San Francisco General Hospital. Our patients were evaluated from April 1985 through December 1987 with an overall mortality rate during this time period of 10070, which is lower than each of these studies. We could detect no difference in mortality for subpopulations analyzed from 1985to 1986and from 1986to 1987.Our lower mortality rate may have reflected earlier disease recognition and initiation of therapy resulting in less severe disease overall, weakening any solitary association between elevated alveolar-arterial oxygen gradient and mortality. Alternatively, a measure of blood oxygen tension or alveolar-arterial oxygen gradient is included in the APACHE II calculation, which may interfere with the ability to distinguish between these two as independent variables in a discriminant analysis. Decreased serum albumin levelhas not been analyzed independently as a predictor of mortality but has been included among several variables associated with poor survival when measured early after hospitalization for first episode PCP (4). Albumin levels were also included as indicators suggesting comparable severity of illness between groups of patients treated with different antipneumocystis drug regimens in one comparative trial (2). Our data substantiated the associa-
BENSON, SPEAR, HINES, POTTAGE, KESSLER, AND TRENHOLME
tion between poor survival and decreased serum albumin level; however, serum albumin was not a statistically significant variable in a stepwise discriminant analysis of variables predicting survival or response to therapy. In addition to their association with mortality, we also evaluated the ability of these four parameters to predict response to initial therapy. Similar to our findings with mortality, only admission APACHE II score and serum LDH were significant in a stepwise discriminant analysis of variables predicting response to therapy. However, the discriminant function with APACHE II and LDH was less successful in correctly grouping patients with regard to response to initial therapy than with regard to survival. One explanation may be that patients were considered to have failed initial therapy based on a clinical judgment after treatment for at least 72 h. It has become apparent with time and increasing experience treating PCP in patients with AIDS, that many do not clinically improve for up to 7 to 10 days after initiation of treatment (5, 8, 9, 14). Because ours was a retrospective analysis, the clinical judgment resulting in a switch to an alternative agent could not be controlled. Although the mean duration of treatment in our study prior to a switch for therapy failure was 7.6 days, some patients received less than 7 days of the initial drug. This may have significantly influenced our analysis of response to initial therapy. In addition, the large number of patients who developed toxicity to the initial drug and were changed to an alternative agent obscured our ability to analyze therapeutic response to a single agent. As applied in this retrospective analysis of our experience, the combination of the APACHE II score and the serum LDH was the most successful of the evaluated parameters in correctly predicting survival for more than 90070 of AIDS patients with first episode PCP. The discriminant function combining the alveolar-arterial oxygen gradient and serum LDH was nearly as successful, correctly predicting survival for more than 89070 of patients. When the alveolar-arterial oxygen gradient wasanalyzed alone using the discriminant function, this parameter correctly classified 86.11 % of patients, suggesting that perhaps this measure is one of the stronger contributing components of the APACHE II score. No other single measurement, analyzed independently or in combination, was more successful than the APACHE II and LDH
in correctly classifying outcome. Based on our observed mortality of 100/0, the predictive capability achieved by the use of these measurements approaches that which we could have achieved by predicting at the outset that all patients would survive. In philosophic terms, wedo "predict" all patients will survive therapy; in practical terms, such assumptions cannot be made. Indirect measures of disease severity may be the best available means to assign specific prognoses to individual patients. A mechanism for prospectively stratifying patients and predicting response to therapy using the APACHE II and LDH may be especially useful when comparing two therapies or when attempting to "match" patients treated with conventional therapies with those treated with experimental therapies in salvage protocols, which may be open and nonrandomized. In this retrospective analysis, we have demonstrated a significant relationship between the combined APACHE II score and serum LDH and two different outcomes, mortality and response to initial therapy, in patients with AIDS and first episode PCP. Their future usefulness as tools for stratifying study patients receiving experimental therapy, for assuring similar severityofillnessof study patients when comparing standard and new therapies, or for predicting response to therapies remains to be substantiated in a prospective manner. Acknowledgment The writers thank Dr. Susan Shott for statistical assistance and Ms. Jacqueline Pawlak for manuscript preparation. References 1. Mills J. Pneumocystiscarinii and Toxoplasma gondii infection in patients with AIDS. Rev Infect Dis 1986; 8:1001-10. 2. Wharton JM, Coleman DL, Wofsy CB, et al. 1Hmethoprim-sulfamethoxazoleor pentamidine for Pneumocystis carinii pneumonia in the acquired immunodeficiency syndrome. Ann Intern Med 1986; 105:37-44. 3. Rainer CA, Feigal DW, Leoung G, et al. Prognosis and natural history of Pneumocystis carinti pneumonia: indicators for early and late survival (abstract). Proceedings of the III International ConferenceonAIDS, Washington, DC 1987. Washington, DC: U.S. Department of Health and Human Services, 1987. THP 154:189. 4. Kales CP, Murren JR, Torres RA, Crocco JA. Early predictors of in-hospital mortality for Pneumocystis cariniipneumonia in the acquired immunodeficiency syndrome. Arch Intern Med 1987; 147:1413-7. 5. Brenner M, Ognibene FP, Lack EE, et al. Prognostic factors and life expectancy of patients with acquired immunodeficiency syndrome and Pneumocystis carinii pneumonia. Am Rev Respir Dis 1987; 136:1199-206.
APACHE II SCORE AND SERUM LDH AS MORTALITY PREDICTORS
6. Medina I, Mills J, Wofsy C. Serum lactate dehydrogenase levels (LDH) in Pneumocystiscarinii pneumonia (PCP) in AIDS: possible indicator and predictor of disease activity (abstract). Proceedings of the III International Conference on AIDS, Washington, DC 1987. Washington, DC: U.S. Department of Health and Human Services, 1987. W.5.5:109. 7. Kagawa KT, Kirsch CM, Yenokida YG, Levine ML. Serum lactate dehydrogenase activity in patients with AIDS and Pneumocystis carinit pneumonia. An adjunct to diagnosis. Chest 1988; 94:1031-33. 8. Sattler FR, Cowan R, Nielson DM, Ruskin J. 'Irimethoprim-sulfamethoxazole compared with pentamidine for treatment of Pneumocystiscarinii pneumonia in the acquired immunodeficiency syn-
drome. A prospective, noncrossover study. Ann Intern Med 1988; 109:280-7. 9. Garay SM, Greene J. Prognostic indicators in the initial presentation of Pneumocystis carinii pneumonia. Chest 1989; 95:769-72. 10. Smith RL, Ripps CS, LewisML. Elevated lactic dehydrogenase values in patients with Pneumocystis carinii pneumonia. Chest 1988; 93:987-92. 11. Zaman MK, White DA. Serum lactate dehydrogenase levels and Pneumocystis cariniipneumonia. Diagnostic and prognostic significance. Am Rev Respir Dis 1988; 137:796-800. 12. Efferen LS, Nadarajah D, Palat DS. Survival following mechanical ventilation for Pneumocystiscariniipneumonia in patients with the acquired immunodeficiency syndrome: a different perspective. Am J Med 1989; 87:401-4.
323 13. Smith RL, Levine SM, Lewis ML. Prognosis of patients with AIDS requiring intensivecare. Chest 1989; 96:857-61. 14. El-Sadr W, Simberkoff MS. Survivaland prognostic factors in severe Pneumocystiscariniipneumonia requiring mechanical ventilation. Am Rev Respir Dis 1988; 137:1264-7. 15. Knaus WA, Draper EA, Wagner DP, et at. APACHE II: a severity of disease classification system. Crit Care Med 1985; 13:818-28. 16. Wagner DP, Knaus WA, Draper EA. Physiologic abnormalities and outcome from acute disease: evidence for a predictable relationship. Arch Intern Med 1986; 146:1389-96. 17. Wagner DP, Draper EA, Campos RA, et at. Initial international use of APACHE. Med Decis Making 1984; 4:297-313.