A proposed scoring system for predicting mortality in melioidosis

TRANSACTIONSOF THE ROYALSOCIETYOF TROPICALMEDICINEAND HYGIENE(2003) 97, 577-581

A proposed scoring system for predicting mortality in melioidosis A l l e n C. C h e n g 1, S u s a n P. J a c u p s 1, Nicholas M. A n s t e y 1'2 and Bart J. C u t t l e 1,z IMenzies School of Health

Research and Northern Territory Clinical School, Flinders University, Casuarina, Australia; 2Royal Darwin Hospital, Darwin, Australia Abstract Melioidosis, due to infection with the environmental organism Burkholderia pseudomallei, continues to be associated with high mortality despite improvements in antibiotic therapy. Using simple clinical findings and baseline laboratory tests available at the time of admission, we attempted to define those patients with acute melioidosis who were at higher risk of death. Using data, collected prospectively from the period October 1989 to June 2002, from patients with acute culture-confirmed melioidosis presenting at the Royal Darwin Hospital, Darwin, Australia, a number of variables were selected that were easily available at the time of admission and reflected organ dysfunction. Mortality was predicted in univariate logistic and multivariate models by the presence of pneumonia, age at diagnosis, serum urea, serum bilirubin, lymphocyte count, and serum bicarbonate. A score was assigned from 0 to 2, based on the degree of abnormality. A melioidosis score was formed from the sum of these scores, with a maximum score of 11. A score of ~< 3 (n = 140) was associated with a mortality of 8.6%, whereas a score of/> 4 (n = 112) was associated with a mortality of 44.6%. Although this scoring system requires external validation, it may help identify a suitable target group of patients for intensive intervention such as early admission to an intensive care unit, the early use of meropenem, and goal-directed resuscitation therapies. Keywords: melioidosis, Burkholderiapseudomallei, prognosis, mortality, multiple organ failure, severity of illness index, Australia Introduction Melioidosis, a disease caused by infection with the Gram-negative bacillus Burkholderia pseudomallei, is endemic in South-East Asia and northern Australia (Dance, 1991). Risk factors for the acquisition of this infection have been clearly defined in both Thai and Australian studies, and include the presence of diabetes, hazardous alcohol use, and chronic renal failure (Suputtamongkol et al., 1994; Currie et al., 2000b). Melioidosis may be disseminated or localized, and both acute and chronic presentations are seen (Currie et al., 2000a). Despite advances in therapy, including the use of ceftazidime and imipenem, acute melioidosis is still associated with a high mortality (Simpson et al., 1999). The ability to identify a high-risk group would aid in the implementation of intensive interventions, such as early antibiotic therapy, early supportive therapy, and experimental immunomodulatory therapies aimed at reducing mortality. Thus, we attempted to define a subgroup of patients presenting with acute melioidosis who were at the highest risk of death. Methods Data on all patients presenting with cultureconfirmed melioidosis to the Royal Darwin Hospital, Darwin, Australia, have been collected prospectively since October 1989. This data includes demographic details, details of the clinical iUness and co-morbidity conditions, and results of radiology and laboratory findings on admission. Ethical approval to review this data was obtained from the H u m a n Research Ethics Committee of the Royal Darwin Hospital and the Menzies School of Health Research, Casuarina, Australia. For the purpose of this analysis, we considered all patients who presented with acute melioidosis confirmed by culture of B. pseudomallei from any site and defined as duration of illness < 2 months (Currie et al., 2000a). T o preserve the assumption of independence of observations, we limited our analysis to first presentations with melioidosis, excluding relapses. Death was Address for correspondence: Allen Cheng, Menzies School of Health Research, P.O. Box 41096, Casuarina, Northern Territory 0811, Australia; phone +61 8 8922 8196, fax +61 8 8927 5187, e-mail [email protected]

defined as mortality attributable to melioidosis and occurring in the context of the acute illness. We examined a number of clinical and simple laboratory parameters available to the clinician at the time of admission (Table 1). The laboratory tests were felt to reflect dysfunction in various organ systems and to be associated with poor outcomes in our clinical experience. Known renal impairment was defined as creatinine > 150 pmol/L prior to the episode of melioidosis. Pneumonia was defined by clinical features suggestive of a lower respiratory tract infection with a new opacity on chest radiograph, confirmed by a positive culture of blood or sputum. Age was taken from the date of admission. If a single biochemical parameter was not assessed on admission, it was assumed to be within the normal range; if > 1 biochemical parameter was missing, the subject was excluded from the analysis. Statistical tests were performed using Intercooled Stata 7.0 for Windows (Stata Corp., College Station, TX, USA). We adopted the following approach: variables found to be significant on univariate logistic regression would be confirmed on multivariate analysis using stepwise selection with a significance level of 0.2. These variables would be examined in greater detail by the use of a generalized additive model (Hastie & Tibshirani, 1990), performed by means of a module formulated by Royston & Ambler (1998). This nonparametric model permits examination of non-linear and threshold effects on survival by use of a scatterplot smoothing function. Break points for a scoring system were based on the normal ranges of the laboratory assays, the shape of the curve on the generalized additive model and the frequency and distribution of abnormal values. A score was formulated by the sum of the variables and plotted on a receiver-operator curve, designed to examine the relationship between sensitivity and specificity. For the purpose of this score, it was felt that sensitivity, the ability to include patients at risk of death, was of greater importance. Finally, the performance of the scoring system was assessed by examining its consistency over the various time periods. Results During the period October 1989 to June 2002, 339 patients presented to the Royal Darwin Hospital with culture-confirmed melioidosis. Fifty-one patients had chronic presentations of melioidosis and were excluded

578

A.C. CHENGETAL.

Table 1. Univariate and multivariate logistic regression of predictors for mortality in patients with acute meHoidosis presenting at the Royal D a r w i n Hospital, D a r w i n , Australia, October 1989-June 2002 Deaths

Total no. of patients

Pneumonia Yes No

50 24

144 144

2.65 (1.48-4.85)

Diabetes Yes No

33 41

118 170

1.22 (0.69-2.15)

Hazardous alcohol intake Yes No

30 44

111 177

1.12 (0.63-1.98)

Known renal impairment Yes No

12 62

27 261

2.56 (1.03-6.21)

Categoricalvariables

Continuous variables Age at admission White cell count Neutrophil count Lymphocyte count Serum urea Serum creatinine Serum bilirubin Serum bicarbonate Serum albumin Serum sodium Serum potassium

Univariate analysis odds ratio (95% CI) 1.04 1.03 1.02 0.47 1.05 1.001 1.03 0.90 0.96 0.94 1.17

(1.02-1.06) (0.99-1.07) (0.98-1.07) (0.31-0.72) (1.02-1.08) (1.001 - 1.003) (1.01-1.04) (0.86-0.94) (0.92-1.002) (0.90-0.98) (0.79-1.74)

Odds ratio (95% CI)

Multivariate analysisa odds ratio (95% CI) 1.07 (1.03-1.10) 0.38 (0.20-0.72) 1.03 (0.99-1.07) 1.02 (1.003-1.04) 0.94 (0.88-1.01)

Multivariate analysis included pneumonia (OR = 2.54, 95 % CI 1.15- 5.61).

from analysis. Analyses were performed on the remaining 288 patients with acute melioidosis. These patients had a median age of 49 years, 73.6% were male, and attributable mortality was 25.7% (n = 74). Mortality decreased over time from 32.7% (35 of 107 patients) during 7 wet seasons between 1989/90 and 1995/96 to 21.7% (39 of 179 patients) between the 1996/97 and 2001/02 wet seasons. The following factors were associated with mortality in the univariate model: pneumonia, baseline renal impairment, age at admission, serum bicarbonate, seru m urea, serum creatinine, lymphocyte count, serum bilirubin, and serum sodium (Table 1). Using both forward and backward stepwise selection, pneumonia, age, serum bicarbonate, serum urea, lymphocyte count, and serum bilirubin were predictive in a multivariate model (Table 1). The relationship of these variables to mortality is illustrated in Fig. 1, with the break points used for the scoring system detailed in Table 2. In formulating the scoring system, 27 patients, with a single biochemical parameter not assessed on admission were included in the analysis; 36 patients with > 1 biochemical parameter not assessed on admission were excluded. The relationship between the total score and mortality is illustrated by the receiver-operator characteristic curve (Fig. 2). The area under this curve was 0.78, indicating fair discriminating ability. An increase in score was associated with a rise in mortality from 6.0% in patients with scores of 0 or 1 to 81.8% in patients with scores of 8 or 9 (Fig. 3). In the group with a total score ~< 3 (n = 140), there were 12 deaths with a total mortality of 8.6% (equivalent to a negative predictive value of 91.4%). Similarly, a total score of i> 4 (n = 112) was associated with a mortality (equivalent to the positive predictive value) of 44.6%. At this cut-off, the sensitivity was 80.6% and

specificity 67.4%, with a positive likelihood ratio of 2.5 and a negative likelihood ratio of 3.4. The mortality in the groups with scores ~< 3 and i> 4 remained relatively constant over the 2 periods detailed above; the positive predictive and negative predictive values during the wet seasons 1989/90 and 1995/96 were 63.9% and 87.5%, and between 1996/97 and 2001/02 they were 35.5% and 94.1%.

Discussion The identification of risk factors in melioidosis has implications for clinical practice, provides an epidemiological tool to compare populations by risk and may offer clues to factors important in pathogenesis. In this observational study using data collected over 12 years, we found 2 clinical features and 4 biochemical parameters predicted mortality. We used these clinical features, the presence of pneumonia and age at diagnosis, and biochemical parameters reflecting acidosis, renal dysfunction, hepatic dysfunction, and lymphopaenia to formulate a scoring system for melioidosis. Other parameters have been shown to be independent predictors of mortality, such as serum interleukin6 and interleukin-10 concentrations (Friedland et al., 1992; Simpson et aL, 2000) and heavy bacteraemia (> 50 colony forming units/mL) measured by pourplate blood cultures (Walsh et al., 1995). Although these assays are not performed routinely in patients with melioidosis, limiting their clinical utility, they provide important insights into pathogenesis. Organ dysfunction has been shown to be predictive of mortality in a variety of other intensive care unit (ICU)-based scoring systems such as the Acute Physiology and Chronic Health Evaluation (APACHE), Multiple Organ Dysfunction Score (MODS), Logistic Organ Dysfunction (LOD) score, and Sequential Organ Failure Assessment (SOFA) score (Pettila et al., 2002). Although our experience has shown that pre-

PROPOSED SCORING SYSTEM FOR MELIOIDOSIS

579

B

A

Y

3-

3-

2

21

1 0

0

-1

--1"

5'0

1'6 2'4 Serum bicarbonate (mmol/L)

6'5 Age (years) D

C 2

0

-1-

-1

Y 1'6 Serum urea (mmol/L)

0:7 113 Lymphocyte count (X 109/L)

E

-1

1'9 3'4 Serum bilirubin (~mol/L) Fig. 1. Generalized additive model plots examining non-linear relationships between continuous variables [(A) age at admission, (B) serum bicarbonate, (C) lymphocyte count, (D) serum urea, and (E) serum bilirubin] and mortality in patients with acute melioidosis (4 d.f., 95% SEM bands).

Table 2. Break p o i n t s for a s c o r i n g s y s t e m to predict m o r t a l i t y in patients with acute m e l i o i d o s i s

Pneumonia Age (years) Serum bicarbonate (mmol/L) Serum urea (mmol/L) S e r u m creatinine (pmol/L) L y m p h o c y t e count (× 109/L) Serum bilirubin 0xmol/L)

0

+1

Absent ~<50 i>24.1 ~<8.0

Present 51-64 16.1-24 8.0-16.0

/>1.3 ~<19

T o t a l score (sum of c o m p o n e n t scores, m a x i m u m score 11)

0.8-1.2 20-33

+2 />65 ~<16 />16.1 />250 ~<0.7 />34

Component scores

A.C. CHENG ETAL.

580

1.00

0.75 0.50 0.25 0.00

0.50

o.25

oSo

o.75

1.oo

1 - specificity Fig. 2. Receiver-operator characteristic curve, examining the sensitivity and specificity of a scoring system for predicting mortality in patients with acute melioidosis. Area under curve = 0.7832.

100 .~ 18I00| 090

'=

patierate ntSIty

-90

MonD . eahtsT°atl

70

t 5o+n 4°tll 6o~

•

-60

"~ E E

-50

~

-40

3°tll

-30

o+1

20 10

0/1

2/3 4/5 6/7 Melioidosis score

8/9

Fig. 3. Distribution of patients, deaths, and mortality rate using a scoring system for predicting mortality in patients with acute melioidosis. Error bars indicate 95% CI.

dicted mortality from APACHE II scores in the small number of patients with septic shock and melioidosis correlated with observed mortality (data not shown), we wished to develop a more simple clinical rule using routine clinical and biochemical parameters. This system is designed for patients with melioidosis generally, rather than in the intensive care setting where complex ICU-based systems may be more appropriate. Chaowagul et al. (1989) previously described similar predictors of mortality in patients with melioidosis in north-east Thailand, where a high mortality was observed in patients with low white cell counts, high urea, hypoglycaemia, and liver dysfunction. However, in this series of 62 patients, limited to those with septicaemia, a detailed analysis of these factors was not attempted (Chaowagul et al., 1989). Within our own population, we have traditionally used the presence of septic shock, defined by standard criteria (ACCP/SCCM, 1992), as a marker of risk and the need for intensive therapies. However, although this group (n = 70) had a mortality of 70%, these patients accounted for only 66% of the mortality associated with melioidosis (data not shown). Of note, factors previously described as being predictive for the acquisition of melioidosis, such as chronic renal failure, alcoholism, and diabetes (Suputtamongkol et al., 1994; Currie et al., 2000b) did not independently predict mortality in this model. We speculate that, although these risk factors are important in the early pathogenesis of melioidosis, the subsequent course of the infection is best reflected by the severity

of organ dysfunction. Whether this is due to bacterial factors such as inoculum dose or virulence of strains, due to host factors such as patterns of cytokine responses or relative neutrophil dysfimction, or due to clinical factors such as delays in commencement of effective antibiotics and other management remains to be defined. The association of lymphopaenia with mortality is intriguing. A small group of patients with melioidosis were shown to have low lymphocyte counts in peripheral blood, but no differences were found between bacteraemic and non-bacteraemic patients. An analysis of lymphocyte subsets found depletion of T cell and natural killer cell subsets, similar to that found after administration of endotoxin (Ramsay et al., 2002). As these cells are important sources of interferon (IF2q)-y in melioidosis (Lertmemongkolchai et aL, 2001) and IFN-y is important in resistance to B. pseudomallei, (Santanirand et al., 1999), it was speculated that these changes may provide conditions conducive to the survival and multiplication ofB. pseudomallei. These parameters, all available at the time of admission, have been combined in a simple predictive model. The composite melioidosis score correlates with mortality. In addition, a cut-off of 3 stratifies patients into 2 groups; low-risk, associated with an overall mortality of < 10%, and high-risk, associated with a mortality of > 40%. We have validated this scoring system internally by assessing its consistency over time; although there has been an overall fall in mortality during this time, mortality remained higher in the high-risk group. There are several limitations to this study. We derived this scoring system on the basis of a relatively low number of cases and validated it in the same data set, and within the limitations of this data. Differences in management may have implications for the generalizability of these findings; thus, this system requires prospective validation in other settings where melioidosis is commonly seen. A Cochrane review of interventions in treating melioidosis suggested that a risk stratification system was required in considering future trials (Samuel and Ti, 2001) and this scoring system may meet this need. This system may also allow for the identification of high-risk patients who may benefit from intensive interventions aimed at reducing mortality, such as the earlier admission to intensive care, the earlier use of meropenem and supportive goal-directed resuscitation therapies (Rivers et al., 2001). Acknowledgements

The authors would like to acknowledge the statistical assistance given by Zhiqiang Wang. A. C. C. is supported by a National Health and Medical Research Council Research Scholarship. References

ACCP/SCCM (1992). American College of Chest Physicians/ Society of Critical Care Medicine Consensus Conference: definitions for sepsis and organ failure and guidelines for the use of innovative therapies in sepsis. Critical Care Medicine, 20, 864-874. Chaowagul, W., White, N. J., Dance, D. A., Wartanagoon, Y., Naigowit, P., Davis, T. M., Looareesuwan, S. & Pitakwatchara, N. (1989). Melioidosis: a major cause of community-acquired septicemia in northeastern Thailand. Journal of Infectious Diseases, 159, 890-899. Currie, B. J., Fisher, D. A., Anstey, N. M. & Jacups, S. P. (2000a). Melioidosis: acute and chronic disease, relapse and re-activation. Transactions of the Royal Society of Tropical Medicine and Hygiene, 94, 301-304. Currie, B. J., Fisher, D. A., Howard, D. M., Burrow, J. N., Lo, D., Selva-Nayagam, S., Anstey, N. M., Huffam, S. E., Snelling, P. L., Marks, P. J., Stephens, D. P., Lure, G. D., Jacups, S. P. & Krause, V. L. (2000b). Endemic melioidosis in tropical northern Australia: a 10-year prospective study and review of the literature. Clinical Infectious Diseases, 31, 981-986.

PROPOSED SCORING SYSTEM FOR MELIOIDOSIS Dance, D. A. (1991). Melioidosis: the tip of the iceberg? Clinical Microbiology Reviews, 4, 52-60. Friedland, J. S., Suputtamongkol, Y., Remick, D. G., Chaowagul, W., Strieter, R. M., Kunkel, S. L., White, N. J. & Griffin, G. E. (1992). Prolonged elevation of interleukin8 and interleukin-6 concentrations in plasma and of leukocyte interleukin-8 mRNA levels during septicemic and localized Pseudomonas pseudomallei infection. Infection and Immunity, 60, 2402-2408. Hastie, T. & Tibshirani, R. (1990). Generalized Additive Models. London: Chapman & Hall. Lertmemongkolchai, G., Cai, G., Hunter, C. A. & Bancroft, G. J. (2001). Bystander activation of CD8(+) T cells contributes to the rapid production of IFN-gamma in response to bacterial pathogens. Journal of Immunology, 166, 1097-1105. Pettila, V., Pettila, M., Sarna, S., Voutilainen, P. & Takkunen, O. (2002). Comparison of multiple organ dysfunction scores in the prediction of hospital mortality in the critically ill. Critical Care Medicine, 30, 1705-1711. Ramsay, S. C., Ketheesan, N., Norton, R., Watson, A. M. & LaBrooy, J. (2002). Peripheral blood lymphocyte subsets in acute human melioidosis. EuropeanJournal of Clinical Microbiology and lnfectious Diseases, 21, 566-568. Rivers, E., Nguyen, B., Havstad, S., Ressler, J., Muzzin, A., Knoblich, B., Peterson, E. & Tomlanovich, M. (2001). Early goal-directed therapy in the treatment of severe sepsis and septic shock. New England Journal of Medicine, 345, 1368-1377. Royston, P. & Ambler, G. (1998). Generalized Additive Models (Stata module STB-42 sg79: gam.ado). London: Royal Postgraduate Medical School, University College London. Samuel, M. & Ti, T. Y. (2001). Interventions for Treating

581

Melioidosis. Oxford: Update Software Ltd, Cochrane Database of Systematic Reviews, No. 2, CD001263. Santanirand, P., Harley, V. S., Dance, D. A., Drasar, B. S. & Bancroft, G. J. (1999). Obligatory role of gamma interferon for host survival in a murine model of infection with Burkholderia pseudomallei. Infection and Immunity, 67, 3593-3600. Simpson, A. J., Suputtamongkol, Y., Smith, M. D., Angus, B. J., Rajanuwong, A., Wuthiekanun, V., Howe, P. A., Walsh, A. L., Chaowagul, W. & White, N. J. (1999). Comparison of imipenem and ceftazidime as therapy for severe melioidosis. Clinical InfectiousDiseases, 29, 381-387. Simpson, A. J., Smith, M. D., Weverling, G. J., Suputtamongkol, Y., Angus, B. ]'., Chaowagul, W., White, N. J., van Deventer, S. J. & Prins, J. M. (2000). Prognostic value of cytokine concentrations (tumor necrosis factor- alpha, interleukin-6, and interleukin-10) and clinical parameters in severe melioidosis. Journal of Infectious Diseases, 181, 621-625. Suputtamongkol, Y., Hall, A. J., Dance, D. A., Chaowagul, W., Rajchanuvong, A., Smith, M. D. & White, N. J. (1994). The epidemiology of melioidosis in Ubon Ratchatani, northeast Thailand. Imernational Journal of Epidemiology, 23, 1082-1090. Walsh, A. L., Smith, M. D., Wuthiekanun, V., Suputtamongkol, Y., Chaowagul, W., Dance, D. A. B., Angus, B. & White, N. J. (1995). Prognostic significance of quantitative bacteremia in septicemic melioidosis. Clinical Infectious Diseases, 21, 1498-1500.

Received 22 November 2002; revised 13 March 2003; accepted for publication 24 March 2003