Experiences of a medical statistician working in a SARS-designated hospital in Singapore

1. Experiences of a medical statistician working in a SARS-designated hospital in Singapore Arul Earnest MSc, DLSHTM, C.Stat School of Public Health, University of Sydney

2. Overview • This presentation will take you through some of the research projects on Severe Acute Respiratory Syndrome (SARS), that were undertaken in Tan Tock Seng Hospital (TTSH) during the recent outbreak in 2003 • I worked in the hospital from 2000 to 2005, providing consultation and collaborating with clinicians from various medical specialities • During the SARS outbreak starting from March 2003, there were several important research projects that originated from TTSH • Let’s look briefly at some of these projects that I have been personally involved in

3. Initial stage of the outbreak • Severe acute respiratory syndrome (SARS) in Singapore: clinical features of index patient and initial contacts. Hsu LY, Lee CC, Green JA, Ang B, Paton NI, Lee L, Villacian JS, Lim PL, Earnest A, Leo YS. Emerg Infect Dis. 2003 Jun;9(6):713-7. • One of the first few case-studies published on SARS world-wide • No fancy statistical methods used, instead the paper made use of simple descriptive results presented in an appropriate and useful manner

4. Initial stage of the outbreak Severe acute respiratory syndrome (SARS) in Singapore: clinical features of index patient and initial contacts. Hsu LY, Lee CC, Green JA, Ang B, Paton NI, Lee L, Villacian JS, Lim PL, Earnest A, Leo YS. Emerg Infect Dis. 2003 Jun;9(6):713-7.

5. Initial stage of the outbreak • Severe acute respiratory syndrome (SARS) in Singapore: clinical features of index patient and initial contacts. Hsu LY, Lee CC, Green JA, Ang B, Paton NI, Lee L, Villacian JS, Lim PL, Earnest A, Leo YS. Emerg Infect Dis. 2003 Jun;9(6):713-7.

6. Initial stage of the outbreak • Severe acute respiratory syndrome (SARS) in Singapore: clinical features of index patient and initial contacts. Hsu LY, Lee CC, Green JA, Ang B, Paton NI, Lee L, Villacian JS, Lim PL, Earnest A, Leo YS. Emerg Infect Dis. 2003 Jun;9(6):713-7. • Conclusions of study • Initial clinical features of SARS are non-specific. • Dry cough is common, although other symptoms of upper respiratory tract infection are unusual. • Chest radiograph may be normal on week 1 of illness. • In early stages, SARS may be hard to differentiate from other viral infections, and diagnostic delays may contribute to the spread of the epidemic.

7. The next paper…in JAMA • Acute respiratory distress syndrome in critically ill patients with severe acute respiratory syndrome. Lew TW, Kwek TK, Tai D, Earnest A, Loo S, Singh K, Kwan KM, Chan Y, Yim CF, Bek SL, Kor AC, Yap WS, Chelliah YR, Lai YC, Goh SK. JAMA. 2003 Jul 16;290(3):374-80. • Informative paper published in Journal of American Medical Association, a journal with a high impact factor • Looked at the determinants of early/ intermediate recovery of SARS patients from the ICU

8. The next paper…in JAMA OBJECTIVE: To describe the clinical spectrum and outcomes of ALI/ARDS in patients with SARS-related critical illness. DESIGN, SETTING, AND PATIENTS: Retrospective case series of adult patients with probable SARS admitted to the intensive care unit (ICU) of a hospital in Singapore between March 6 and June 6, 2003. MAIN OUTCOME MEASURES: The primary outcome measure was 28-day mortality after symptom onset.

9. The next paper…in JAMA The characteristics of ICU and non-ICU SARS patients were different: ICU patients were more likely to be older, less likely to be healthcare workers, likely to have shortness of breath, be on corticosteroids and have several pre-existing comorbidities

10. The next paper…in JAMA

11. The next paper…in JAMA • RESULTS: • Of 199 patients hospitalized with SARS, 46 (23%) were admitted to the ICU • Mortality at 28 days for the entire cohort was 20 (10.1%) of 199 and for ICU patients was 17 (37%) of 46. • Lower Acute Physiology and Chronic Health Evaluation II scores and higher baseline ratios of PaO2 to fraction of inspired oxygen were associated with earlier recovery. • CONCLUSIONS: • Critically ill patients with SARS and ALI/ARDS had characteristic clinical findings, high rates of complications; and high mortality.

12. Predictors of mortality among ICU patients Clinical features and predictors for mortality in a designated national SARS ICU in Singapore. T ai DY, Lew TW, Loo S, Earnest A, Chen MI; Tan Tock Seng Hospital SARS ICU Group. Ann Acad Med Singapore. 2003 Sep;32(5 Suppl):S34-6. This time, the study examined the factors associated with mortality among ICU patients with SARS, using survival analysis techniques. i.e. the Cox regression model.

13. Predictors of mortality among ICU patients • Clinical features and predictors for mortality in a designated national SARS ICU in Singapore. T ai DY, Lew TW, Loo S, Earnest A, Chen MI; Tan Tock Seng Hospital SARS ICU Group. Ann Acad Med Singapore. 2003 Sep;32(5 Suppl):S34-6. • Methods: • Survival time was defined as the interval between date of illness and death or censoring as of 9 June 2003. • Following risk factors were studied: sex, age, race, co-morbidities, PaO2/FiO2 ratio on admission to ICU, lowest PaO2/FiO2 ratio, Acute Physiology and Chronic Health Evaluation (APACHE) II score, highest serum lactate dehydrogenase (LDH), lowest serum sodium, lowest serum potassium, lowest absolute lymphocyte count (ALC), highest total white count (TWC), highest prothrombin time (PT) and highest activated partial thromboplastin time (PTT). • Univariate Cox regression models were fitted for each of the variables.

14. Predictors of mortality among ICU patients • Clinical features and predictors for mortality in a designated national SARS ICU in Singapore. T ai DY, Lew TW, Loo S, Earnest A, Chen MI; Tan Tock Seng Hospital SARS ICU Group. Ann Acad Med Singapore. 2003 Sep;32(5 Suppl):S34-6. • Methods: • Starting from the most significant, variables that were significant in the univariate analysis were included in the multivariate model in turn. • Likelihood ratio test was used to test whether inclusion of a new covariate helped improve the fit of the model. • The Nelson-Aalen cumulative hazard plot (on the log scale) was used to graphically examine the proportional hazard assumption. • The Schoenfeld test was used to formally ascertain if the assumption of proportional hazards was violated for the final model.

15. Predictors of mortality among ICU patients Clinical features and predictors for mortality in a designated national SARS ICU in Singapore. T ai DY, Lew TW, Loo S, Earnest A, Chen MI; Tan Tock Seng Hospital SARS ICU Group. Ann Acad Med Singapore. 2003 Sep;32(5 Suppl):S34-6. Conclusion: About one in five probable SARS patients required ICU care. This group of critically ill SARS patients has high mortality and morbidity. The predictors for ICU mortality were male gender, APACHE II score > 15 and history of congestive cardiac failure. Wide CIs probably due to small number of CCF comorbidities

16. Evaluating the effectiveness of ribavirin • Investigational use of ribavirin in the treatment of severe acute respiratory syndrome, Singapore, 2003. Leong HN, Ang B, Earnest A, Teoh C, Xu W, Leo YS. Trop Med Int Health. 2004 Aug;9(8):923-7. Objective: Ribavirin is a broad spectrum nucleoside analogue efficacious in the treatment of several viral infections. In the recent severe acute respiratory syndrome (SARS) outbreak, ribavirin was used in various countries against this novel coronavirus. We conducted a retrospective analysis to assess the efficacy of ribavirin in the treatment of SARS in Singapore. • Methods: • A total of 229 cases were analysed. • Ninety-seven (42.4%) patients received ribavirin at a mean of 6.4 days of illness. • Univariate analysis using Fisher’s exact test and Student’s t-test. • Multivariate analysis was performed using Cox regression model with death as the outcome of interest.

17. Evaluating the effectiveness of ribavirin Ribavirin prescribed in early stage of outbreak

18. Evaluating the effectiveness of ribavirin Some clinical and demographic differences found between those prescribed ribavirin and those who were not. This is not surprising as the study was not an RCT

19. Evaluating the effectiveness of ribavirin Modelling was done sequentially to highlight the effect of the measured confounders on ribavirin’s effect on mortality. As we can see, there is a reversal in the direction of the effect of ribavirin, but the results were not significant, even though the numbers were not too small.

20. Evaluating the effectiveness of ribavirin • Investigational use of ribavirin in the treatment of severe acute respiratory syndrome, Singapore, 2003. Leong HN, Ang B, Earnest A, Teoh C, Xu W, Leo YS. Trop Med Int Health. 2004 Aug;9(8):923-7.Results: • The treatment group had younger women with more symptoms of myalgia (P < 0.001). • The crude death rate was 12.9% in the control and 10.3% (P=0.679) in the treatment group. After correction for age, male sex, lactate dehydrogenase levels and steroid use, the hazard ratio was 1.03 (95% CI: 0.44–2.41, P =0.939). • Conclusion: • In this retrospective, uncontrolled cohort analysis, use of ribavirin did not appear to confer any benefit for patients with SARS.

21. Lung function among survivors of SARS • Pulmonary function and exercise capacity in survivors of severe acute respiratory syndrome. Ong KC, Ng AW, Lee LS, Kaw G, Kwek SK, Leow MK, Earnest A.Eur Respir J. 2004 Sep;24(3):436-42. Aims: • To investigate pulmonary function and exercise capacity in a group of survivors of the severe acute respiratory syndrome (SARS). • At 3 months after hospital discharge, 46 survivors of SARS underwent the following evaluation: spirometry, static lung volumes and carbon monoxide transfer factor (TL,CO). • In total, 44 of these patients underwent cardiopulmonary exercise testing.

22. Lung function among survivors of SARS • Pulmonary function and exercise capacity in survivors of severe acute respiratory syndrome. Ong KC, Ng AW, Lee LS, Kaw G, Kwek SK, Leow MK, Earnest A.Eur Respir J. 2004 Sep;24(3):436-42. Results: • No abnormalities were detected in the pulmonary function tests in 23 (50%) of the patients. • Abnormalities of forced vital capacity (FVC), forced expiratory volume in one second (FEV1), FEV1/FVC and TL,CO were detected in seven (15%), 12 (26%), one (2%) and 18 (39%) patients, respectively. • All of these abnormalities were mild except in one case. • In 18 patients (41%), the maximum aerobic capacity was below the lower limit of the normal range. Breathing reserve was low in four patients and significant oxygen desaturation was detected in a further four patients.

23. Haematological parameter changes in severe acute respiratory syndrome patients. Haematological parameters in severe acute respiratory syndrome. Chng WJ, Lai HC, Earnest A, Kuperan P. Clin Lab Haematol. 2005 Feb;27(1):15-20. Aims: We studied changes in haematological parameters in SARS patients using median values analysed according to the day of symptom onset. White cell (WCC), absolute neutrophil, absolute lymphocyte (ALC) and platelet counts followed a v-shaped trend with the nadir at day 6 or 7 after symptom onset except for ALC in the ICU group that had not reached the nadir by day 12.

24. Haematological parameter changes in severe acute respiratory syndrome patients. By plotting the median values for each haematological parameter over a period of 12 days from symptom onset, time trends for changes to Hb, WCC, platelet count, ANC and ALC were obtained (Figure 1). The Hb for both groups continued to fall during the first 12 days of illness (graph not shown). The WCC initially decreased, reaching a nadir at day 7 or 8 of illness. The median nadir WCC was 4 x 10(9(/l (range 2–12 x 10(9)/l for non-ICU and 2–14.5 x 10(9)/l for ICU group) for both groups. Changes in the ANC reflected changes in the WCC. The median nadir ANC was 2.66 x 10(9)/l (range 1.08–8.85 x 10(9)/l) for the non-ICU group and higher for the ICU group at 2.8 x 10(9)/l (range 1.29–12.7 x 10(9)/l).

25. Differentiating SARS and Dengue • Use of simple laboratory features to distinguish the early stage of severe acute respiratory syndrome from dengue fever. Wilder-Smith A, Earnest A, Paton NI. Clin Infect Dis. 2004 Dec 15;39(12):1818-23. Epub 2004 Nov 19. • BACKGROUND: • The diagnosis of severe acute respiratory syndrome (SARS) is difficult early in the illness, because its presentation resembles that of other non-specific viral fevers, such as dengue. • Dengue fever is endemic in many of the countries in which the large SARS outbreaks occurred in early 2003. Misdiagnosis may have serious public health consequences. • We aimed to determine simple laboratory features to differentiate SARS from dengue. • METHODS: • We compared the laboratory features of 55 adult patients with SARS at presentation (who were all admitted before radiological changes had occurred) and 147 patients with dengue. • Features independently predictive of dengue were modeled by multivariate logistic regression to create a diagnostic tool with 100% specificity for dengue.

26. Differentiating SARS and Dengue • Use of simple laboratory features to distinguish the early stage of severe acute respiratory syndrome from dengue fever. Wilder-Smith A, Earnest A, Paton NI. Clin Infect Dis. 2004 Dec 15;39(12):1818-23. Epub 2004 Nov 19. • RESULTS: • Multivariate analysis identified 3 laboratory features that together are highly predictive of a diagnosis of dengue and able to rule out the possibility of SARS • Platelet count of <140 x 10(9) platelets/L, white blood cell count of <5x10(9) cells/L, and aspartate aminotransferase level of >34 IU/L. • A combination of these parameters has a sensitivity of 75% and a specificity of 100%. • CONCLUSIONS: • Simple laboratory data may be helpful for the diagnosis of disease in adults admitted because of fever in areas in which dengue is endemic when the diagnosis of SARS needs to be excluded. • Application of this information may help to optimize the use of isolation rooms for patients presenting with nonspecific fever.

27. Differentiating SARS and Dengue Several covariates were identified as significantly different between dengue and SARS in the univariate analysis. However, many of these variables are related to one another (confounders)

28. Differentiating SARS and Dengue A model with the combination of these 3 parameters and using these cutoff values identifies dengue correctly in 75% (sensitivity of 75%) and rules out dengue in 100% (specificity of 100%) of cases. From a public health point of view, a specificity of 100% is desirable so that no case of SARS will be misdiagnosed as dengue, and thus, the patient not be isolated, leading to secondary transmission.

29. Modelling bed-occupancy during SARS • Using autoregressive integrated moving average (ARIMA) models to predict and monitor the number of beds occupied during a SARS outbreak in a tertiary hospital in Singapore. Earnest A, Chen MI, Ng D, Sin LY. BMC Health Serv Res. 2005 May 11;5(1):36. • BACKGROUND: • Apply autoregressive integrated moving average (ARIMA) models to make real-time predictions on the number of beds occupied in Tan Tock Seng Hospital, during the recent SARS outbreak. • METHODS: • This is a retrospective study design. • Hospital admission and occupancy data for isolation beds was collected from Tan Tock Seng hospital for the period 14th March 2003 to 31st May 2003. • The main outcome measure was daily number of isolation beds occupied by SARS patients.

30. Modelling bed-occupancy during SARS • Using autoregressive integrated moving average (ARIMA) models to predict and monitor the number of beds occupied during a SARS outbreak in a tertiary hospital in Singapore. Earnest A, Chen MI, Ng D, Sin LY. BMC Health Serv Res. 2005 May 11;5(1):36. • METHODS: • Among the covariates considered were daily number of people screened, daily number of people admitted (including observation, suspect and probable cases) and days from the most recent significant event discovery. • We utilized the following strategy for the analysis. • Firstly, we split the outbreak data into two. • Data from 14th March to 21st April 2003 was used for model development. • We used ARIMA models in an attempt to model the number of beds occupied.

31. Modelling bed-occupancy during SARS

32. Modelling bed-occupancy during SARS

33. Modelling bed-occupancy during SARS • Using autoregressive integrated moving average (ARIMA) models to predict and monitor the number of beds occupied during a SARS outbreak in a tertiary hospital in Singapore. Earnest A, Chen MI, Ng D, Sin LY. BMC Health Serv Res. 2005 May 11;5(1):36. • RESULTS: • We found that the ARIMA (1,0,3) model was able to describe and predict the number of beds occupied during the SARS outbreak well. • The mean absolute percentage error (MAPE) for the training set and validation set were 5.7% and 8.6% respectively • Furthermore, the model also provided three-day forecasts of the number of beds required. • Total number of admissions and probable cases admitted on the previous day were also found to be independent prognostic factors of bed occupancy. • CONCLUSION: ARIMA models provide useful tools for administrators and clinicians in planning for real-time bed capacity during an outbreak of an infectious disease such as SARS. The model could well be used in planning for bed-capacity during outbreaks of other infectious diseases as well.

34. Asymptomatic SARS infection Asymptomatic SARS coronavirus infection among healthcare workers, Singapore. Wilder-Smith A, Teleman MD, Heng BH, Earnest A, Ling AE, Leo YS. Emerg Infect Dis. 2005 Jul;11(7):1142-5. We conducted a study among healthcare workers (HCWs) exposed to patients with severe acute respiratory syndrome (SARS) before infection control measures were instituted. Of all exposed HCWs, 7.5% had asymptomatic SARS-positive cases. Asymptomatic SARS was associated with lower SARS antibody titers and higher use of masks when compared to pneumonic SARS.

35. Asymptomatic SARS infection Asymptomatic SARS coronavirus infection among healthcare workers, Singapore. Wilder-Smith A, Teleman MD, Heng BH, Earnest A, Ling AE, Leo YS. Emerg Infect Dis. 2005 Jul;11(7):1142-5.

36. Asymptomatic SARS infection Asymptomatic SARS coronavirus infection among healthcare workers, Singapore. Wilder-Smith A, Teleman MD, Heng BH, Earnest A, Ling AE, Leo YS. Emerg Infect Dis. 2005 Jul;11(7):1142-5.

37. Conclusions • There were several important clinical and public health questions about SARS that has been answered through research projects in TTSH • Ample opportunities for biostatisticians to do research in a clinical setting during an outbreak of an infectious disease • Proximity to front-line clinicians helps promote research • Challenges in working with specialists from various fields and in a time-pressed environment. i.e. need answers yesterday • A biostatistician often facilitates multi-disciplinary research (e.g. working with an anaesthetist, respiratory and infectious disease specialist together can spark research ideas)