Serial Evaluation of the SOFA Score to Predict Outcome in Critically Ill Patients

CARING FOR THE CRITICALLY ILL PATIENT

Flavio Lopes Ferreira, MD Context Evaluation of trends in organ dysfunction in critically ill patients may help Daliana Peres Bota, MD predict outcome. Annette Bross, MD Objective To determine the usefulness of repeated measurement the Sequential Or- gan Failure Assessment (SOFA) score for prediction of mortality in intensive care unit Christian Me´lot, MD, PhD, (ICU) patients. MSciBiostat Design Prospective, observational cohort study conducted from April 1 to July 31, Jean-Louis Vincent, MD, PhD 1999. Setting A 31-bed medicosurgical ICU at a university hospital in Belgium. UTCOME PREDICTION IS IM- Patients Three hundred fifty-two consecutive patients (mean age, 59 years) admit- portant both in clinical and ted to the ICU for more than 24 hours for whom the SOFA score was calculated on administrative intensive admission and every 48 hours until discharge. care unit (ICU) manage- ⌬ ment.O1 Although outcome prediction Main Outcome Measures Initial SOFA score (0-24), -SOFA scores (differences between subsequent scores), and the highest and mean SOFA scores obtained during and measurement should not be the the ICU stay and their correlations with mortality. only measure of ICU performance, outcome prediction can be usefully ap- Results The initial, highest, and mean SOFA scores correlated well with mortality. Initial and highest scores of more than 11 or mean scores of more than 5 corre- plied to monitor the performance of an sponded to mortality of more than 80%. The predictive value of the mean score was individual ICU and possibly to com- independent of the length of ICU stay. In univariate analysis, mean and highest SOFA pare ICUs. Outcome prediction can also scores had the strongest correlation with mortality, followed by ⌬-SOFA and initial be useful in providing information on SOFA scores. The area under the receiver operating characteristic curve was largest likely patient outcomes for relatives of for highest scores (0.90; SE, 0.02; PϽ.001 vs initial score). When analyzing trends in critically ill patients and potentially for the SOFA score during the first 96 hours, regardless of the initial score, the mortality therapeutic decision making and guid- rate was at least 50% when the score increased, 27% to 35% when it remained un- ing resource allocation. Outcome pre- changed, and less than 27% when it decreased. Differences in mortality were better diction models currently available have predicted in the first 48 hours than in the subsequent 48 hours. There was no significant difference in the length of stay among these groups. Except for initial scores of not been validated for use in directing more than 11 (mortality rate Ͼ90%), a decreasing score during the first 48 hours was 2 individual patient management. associated with a mortality rate of less than 6%, while an unchanged or increasing Currently available outcome predic- score was associated with a mortality rate of 37% when the initial score was 2 to 7 tion models (such as the APACHE and 60% when the initial score was 8 to 11. [Acute Physiology and Chronic Health Conclusions Sequential assessment of organ dysfunction during the first few days 3 Evaluation], SAPS [simplified acute of ICU admission is a good indicator of prognosis. Both the mean and highest SOFA physiology score],4 and MPM [mortal- scores are particularly useful predictors of outcome. Independent of the initial score, ity probability models]5 systems) cal- an increase in SOFA score during the first 48 hours in the ICU predicts a mortality rate culate a prediction on values taken of at least 50%. within the first 24 hours of an ICU stay. JAMA. 2001;286:1754-1758 www.jama.com However these scores ignore the many 7 Author Affiliation: Department of Intensive Care, factors that can influence patient out- proportion of the ICU budget. Re- Erasme University Hospital, Free University of Brus- come during the course of an ICU stay. cently developed organ failure scores, sels, Belgium. Corresponding Author and Reprints: Jean-Louis Vin- Being able to evaluate changes in pa- such as the Sequential Organ Failure As- cent, MD, PhD, Department of Intensive Care, Erasme tient status over time thus represents sessment (SOFA) (TABLE 1)8 can help as- University Hospital, Route de Lennik, 808, B-1070 Brus- sels, Belgium (e-mail: [email protected]). an improvement on standard models. sess organ dysfunction or failure over Caring for the Critically Ill Patient Section Editor: Debo- Organ dysfunction is associated with time and are useful to evaluate morbid- rah J. Cook, MD, Consulting Editor, JAMA. Advisory Board: David Bihari, MD; Christian Brun- high rates of ICU morbidity and mortality. Although these scoring systems were Buisson, MD; Timothy Evans, MD; John Heffner, MD; 6,7 ity, and, as such, accounts for a high developed to describe and quantify or- Norman Paradis, MD.

Downloaded From: https://jamanetwork.com/ on 09/25/2021 SOFA SCORE TO PREDICT OUTCOME

Table 1. The Sequential Organ Failure Assessment (SOFA) Score* SOFA Score

Variables 012 3 4 Respiratory Pao2/FIO2,mmHg Ͼ400 Յ400 Յ300 Յ200† Յ100† Coagulation Platelets ϫ103/µL‡ Ͼ150 Յ150 Յ100 Յ50 Յ20 Liver Bilirubin, mg/dL‡ Ͻ1.2 1.2-1.9 2.0-5.9 6.0-11.9 Ͼ12.0 Cardiovascular Hypotension No hypotension Mean arterial Dop Յ5ordob Dop Ͼ5, epi Յ0.1, Dop Ͼ15, epi Ͼ0.1, pressure (any dose)§ or norepi Յ0.1§ or norepi Ͼ0.1§ Ͻ70 mm Hg Central nervous system Glasgow Coma Score Scale 15 13-14 10-12 6-9 Ͻ6 Renal Creatinine, mg/dL Ͻ1.2 1.2-1.9 2.0-3.4 3.5-4.9 or Ͻ500 Ͼ5.0 or Ͻ200 or urine output, mL/d࿣

*Norepi indicates norepinephrine; Dob, dobutamine; Dop, dopamine; Epi, epinephrine; and FIO2, fraction of inspired oxygen. †Values are with respiratory support. ‡To convert bilirubin from mg/dL to µmol/L, multiply by 17.1. §Adrenergic agents administered for at least 1 hour (doses given are in µg/kg per minute). ࿣To convert creatinine from mg/dL to µmol/L, multiply by 88.4.

gan function and not to predict out- dated patients, the assumed Glasgow Table 2. Demographics of Study come, the obvious relationship be- Coma Score Scale was used to evalu- Population* tween organ dysfunction and mortality ate the neurological status. Characteristics Values has been demonstrated in several stud- The total SOFA was calculated as the No. of patients 352 ies.9-12 We were interested in evaluating sum of all daily SOFA scores during the Age, mean (SD) [range], y 59 (17) [18-95] Sex whether repeated measurement of the ICU stay for each patient. The mean score Men 230 SOFA score, by including alterations over was defined as the ratio of total score to Women 122 time, could help refine outcome predic- the length of stay (LOS) in the ICU. The Type of admission, No. (%) Medical 195 (55.4) tion. highest score recorded during the ICU Surgical 157 (44.6) stay was also noted. The ⌬-SOFA score Length of ICU stay, d METHODS Mean 6.5 was defined as the difference between 2 Median 4.0 Following approval by the the ethical subsequent scores; for example, the Range 1-56 review board of Erasme University Hos- ⌬-SOFA score 48-0 was the difference be- No. (%) of deaths 81 (23) pital, Free University of Brussels, Bel- tween the 48-hour SOFA score and the *ICU indicates intensive care unit. gium, which waived informed con- admission score. sent on the basis that this was an Odds ratios with 95% confidence in- RESULTS epidemiological study without inter- tervals were computed using a univari- The study included 352 patients with a vention, all patients (Ͼ18 years) ad- ate logistic regression model with ICU mean (SD) age of 59 (17) years mitted to the 31-bed medicosurgical de- outcome as the dependent variable. A (TABLE 2). From the expected 13620 partment of intensive care for more than ␹2 statistics test (with Yates correction variables, 267 were missing (215 bili- 24 hours during a 4-month period when applicable) was used to evaluate rubin levels, 21 creatinine concentra- (April 1-July 31, 1999) were included the statistical significance of categori- tions, 15 PaO2/FIO2 [fraction of inspired in the study. cal variables. The results are pre- oxygen] ratios, and 16 platelet counts). Demographic, laboratory, and clini- sented as mean (SD). Comparisons of As expected, the initial SOFA score was cal data were collected, and the SOFA the areas under the receiver operating significantly related to vital status. An score (0-24, Table 1) was calculated, on characteristic (ROC) curves were also initial SOFA score up to 9 predicted a admission and every 48 hours until dis- performed with a test based on the dif- mortality of less than 33% while an charge. In the calculation of the score, ference between the 2 areas and the SE initial SOFA score of greater than 11 the worst values for each parameter in of the difference.13 Using Statview (SAS predicted a mortality rate of 95% the 24-hour period were used. For a Institute, Cary, NC) and Medcal (Med- (FIGURE 1A). The highest SOFA score single missing value, a replacement was cal Software, Mariakerke, Belgium), was also correlated with mortality: high- calculated from the mean of the sum of all statistical tests were 2-tailed and a est scores of 10 correlated with a mor- the results immediately preceding and P value Ͻ.05 was considered signi- tality rate of 40% while those higher than following the missing value. In se- ficant. 11 were associated with a mortality rate

Downloaded From: https://jamanetwork.com/ on 09/25/2021 SOFA SCORE TO PREDICT OUTCOME

greater than 80% (Figure 1B). The mean tial SOFA scores ranged from 8 to 11, and Figure 1. Mortality Rate in Relation to the Changes in Sequential Organ Failure SOFA score over the entire ICU stay was 91% when the initial SOFA score was Assessment (SOFA) Score During the First also correlated with mortality (Figure higher than 11. 48 Hours in the Intensive Care Unit 1C). The predictive value of the mean SOFA score for mortality was similar COMMENT A Initial SOFA Score In developing a scoring system, such as 100 20 regardless of the LOS. By univariate logistic analysis, the SOFA, for assessing and monitoring or- 80 mean SOFA score correlated most gan dysfunction, several important fea-

60 closely with mortality (TABLE 3), fol- tures need to be considered. First, or- 12 lowed by the highest score, the ⌬-SOFA gan failure is not an all-or-nothing 40 11 48-0 score, and the initial score. The phenomenon; rather, it is a con- 14 Mortality Rate, % 20 highest SOFA score presented the larg- tinuum of alterations in organ func- 5 18 0 est area under the ROC curve (0.90, SE tion from normal function, through 0 0-1 2-3 4-5 6-7 8-9 10-11 >11 0.02) compared with the other SOFA- varying degrees of dysfunction, to or- No. of gan failure. Second, the description of Patients 43 77 8965 33 24 21 derived variables, followed by the mean SOFA score (area under ROC curve organ dysfunction needs to be based on B Highest SOFA Score 0.88, SE 0.03). The area under the ROC simple, easily repeatable variables spe- 100 26 curve was significantly larger for the cific to the organ in question and readily 16 80 highest SOFA score than for the initial available in all institutions. Third, or- SOFA score (PϽ.001, FIGURE 2). gan dysfunction is not static. It will al- 60 11 Trends in SOFA scores during the first ter over time, and a scoring system 40 48 hours were also analyzed. Regard- needs to be able to take this time fac- 10 tor into account. In using the SOFA for Mortality Rate, % 12 20 less of the initial score, the mortality rate 5 was 50% or higher when the score in- outcome prediction, the ability to per- 0 1 0 creased, 27% to 35% when it did not form serial SOFA scores allow a more 0-1 2-3 4-5 6-7 8-9 10-11 12-14 >14 No. of change, and less than 27% when it de- effective representation of the dynam- Patients 33 6775 66 38 24 20 29 creased (TABLE 4). Differences in mor- ics of illness including the effects of C Mean SOFA Score tality were predicted better during the therapy compared with traditional out- 100 first 48 hours than in the subsequent 48 come prediction models at the time of 27 ICU admission. Although some inves- 80 19 hours. There was no significant difference in LOS among these groups. When tigators have used the APACHE II score 60 we analyzed this trend, taking into ac- over time,14-16 this process has never 40 13 count the initial SOFA score for values been validated. Derived measures from

16 of 11 or lower, a decreasing value was the APACHE III system have also been Mortality Rate, % 20 proposed for use on a daily basis,17 but 5 associated with a mortality rate of 6% or 1 0 less (FIGURE 3). However, when the APACHE III is not available in the pub- 0-1 1.1-2 2.1-3 3.1-4 4.1-5 >5.1 SOFA Score mean SOFA score increased or re- lic domain, and its daily use has again No. of mained unchanged, the mortality rate av- not been validated. Patients 85 93 80 36 26 32 eraged 37% when the initial SOFA scores The SOFA score is a useful tool to Numbers above the bars indicate number of deaths. ranged from 2 to 7, 60% when the ini- stratify and compare patients in clinical trials.18,19 Moreno et al12 recently Table 3. Univariate Logistic Regression Analysis of Length of Stay and Sequential Organ demonstrated that the initial SOFA Failure Assessment (SOFA) Derived Parameters as Predictors of Mortality score can be used to quantify the de- Odds Ratio gree of organ dysfunction or failure pres- Variables Coefficient, Mean (SE) (95% Confidence Interval) P Value ent on admission, that the ⌬-SOFA Mean SOFA score 1.12 (0.13) 3.06 (2.36-3.97) Ͻ.001 score can demonstrate the degree of Ͻ Highest SOFA score 0.46 (0.05) 1.59 (1.43-1.76) .001 dysfunction or failure developing dur- ⌬ * Ͻ -SOFA score, 48-0 0.42 (0.08) 1.52 (1.29-1.78) .001 ing an ICU stay, and that the total maxi- Initial SOFA score 0.37 (0.05) 1.45 (1.32-1.59) Ͻ.001 SOFA score at 48 h 0.37 (0.06) 1.45 (1.30-1.61) Ͻ.001 mum SOFA score can represent the cu- SOFA score at 96 h 0.33 (0.06) 1.39 (1.22-1.57) Ͻ.001 mulative organ dysfunction experienced ⌬-SOFA score, 96-0† 0.21 (0.09) 1.24 (1.04-1.47) .02 by the patient. They also demon- Length of stay 0.07 (0.02) 1.07 (1.03-1.11) Ͻ.001 strated a strong correlation of all these Total SOFA score 0.04 (0.01) 1.05 (1.03-1.06) Ͻ.001 parameters with mortality outcome. *Represents the difference between the 48-hour SOFA score and the admission score. In our study, we have moved a step †Represents the difference between the 96-hour SOFA score and the admission score. further, presenting selected SOFA

Downloaded From: https://jamanetwork.com/ on 09/25/2021 SOFA SCORE TO PREDICT OUTCOME

parameters, the mean and the highest Figure 2. Comparisons of the Areas Under the Receiver Operating Characteristic (ROC) SOFA scores, as reliable predictors of Curves for Prediction of Mortality outcome throughout the ICU stay. The mean SOFA score gives an indication of A Initial SOFA Score B Mean SOFA Score the average degree of organ failure over 100 100 time and could also be a useful tool for 80 80 stratifying patients in clinical trials, ac- (>2) cording to the total score or the scores 60 60 for individual organs. The highest SOFA (>8) score can identify the critical point at 40 40 which patients exhibit the highest de- % Sensitivity, gree of organ dysfunction during their 20 20 ICU stay. With these 2 variables, we can 0.79 (0.75-0.83) 0.88 (0.84-0.92) thus define the peak and the total 0 0 0 2040 60 80 100 0 2040 60 80 100 amount of organ impairment for any patient or group of patients during their C Highest SOFA Score D Total SOFA Score ICU stay. The equivalence of the areas 100 100 under the ROC curve for these 2 parameters suggest that they are similarly ef- 80 80 (>13) (>8) fective in predicting outcome. 60 60 The ⌬-SOFA score could be used to reflect patient response to therapeutic 40 40 strategies and allow the physician to % Sensitivity, monitor daily progress, offering an 20 20 objective evaluation treatment re- 0.90 (0.86-0.93) 0.85 (0.81-0.89) sponses. For example, knowledge of the 0 0 0 2040 60 80 100 0 2040 60 80 100 trend in SOFA score over time could fa- cilitate decision making regarding the E SOFA Score at 48 h F SOFA Score at 96 h appropriateness of instituting organ 100 100 support. Knowing that a decreasing (>4) SOFA score is associated with an im- 80 80

proved outcome should prompt aggres- 60 (>7) 60 sive early therapy, which may reduce 20 mortality. Others have shown that the 40 40 development of organ failure may oc- % Sensitivity, cur early during an ICU stay,21 and a 20 20 scoring system that allows regular sur- 0.84 (0.78-0.88) 0.82 (0.74-0.88) 0 0 veillance of organ function is thus 0 2040 60 80 100 0 2040 60 80 100 needed. Trends in the SOFA score over ∆ ∆ the first 48 hours of an ICU stay could G -SOFA Score 48-0 h H -SOFA Score 96-0 h provide such a system and be a sensi- 100 100 tive indicator of outcome, as reflected 80 80 in the fact that a decreasing value was (>0) associated with a decrease in mortal- 60 60 ity rates from 50% to 27%. Interestingly, the LOS was not re- 40 40 Sensitivity, % Sensitivity, lated to outcome prediction. Indeed, the (>1) mean SOFA score had a better prognos- 20 20 tic value than the other SOFA derived 0.76 (0.69-0.82) 0.62 (0.53-0.70) 0 0 variables. This may be because patients 0 2040 60 80 100 0 2040 60 80 100 who present with a limited degree of or- 100–Specificity, % 100–Specificity, % gan dysfunction and have a long ICU stay The area and the 95% confidence interval are presented in each panel. The ⌬ scores represent difference be- still have a high likelihood of survival. tween the 48-hour SOFA score and the admission score and the difference between the 96-hour SOFA score In conclusion, evaluation of the and the admission score. Data markers are the optimal threshold for each SOFA score that discriminates be- SOFA score throughout the ICU stay tween survival and nonsurvival.

Downloaded From: https://jamanetwork.com/ on 09/25/2021 SOFA SCORE TO PREDICT OUTCOME

Table 4. Changes in Sequential Organ Failure Assessment (SOFA) Score in Relation to Outcome* First 48 Hours Next 48 Hours

No. at % of Deaths Mean LOS No. at %of Total % Average % of Deaths Evolution Risk (SE) 95% CI (SE) Evolution Risk Deaths of Deaths Over First 96 h Increased 66 53 (6.1) 41-65 12.4 (6.4) Increased 22 67 57 Unchanged 12 5 53 Ͼ50 Decreased 32 44 50 Unchanged 32 31 (8.2) 15-47 12.6 (9.5) Increased 9 33 32 Unchanged 9 11 27 27-35 Decreased 14 43 35 Decreased 30 23 (7.7)† 8-38 10.9 (9.3) Increased 12 33 26 Unchanged 6 33 25 Ͻ27 Decreased 12 8 19 *LOS indicates length of stay. †P = .01, increased vs decreased.

Statistical expertise: Me´lot. Use of the SOFA score to assess the incidence of or- Figure 3. Relation Between Mortality Rates Obtained funding: Vincent. gan dysfunction/failure in intensive care units: re- and Sequential Organ Failure Assessment Administrative, technical, or material support: sults of a multicentric, prospective study. Crit Care Med. (SOFA) Scores Vincent. 1998;26:1793-1800. Study supervision: Vincent. 11. Antonelli M, Moreno R, Vincent JL, et al. Appli- cation of SOFA score to trauma patients: Sequential Change in SOFA Score During First 48 Hours Organ Failure Assessment. Intensive Care Med. 1999; Decrease Increase or No Change REFERENCES 25:389-394. 12. Moreno R, Vincent JL, Matos A, et al. The use of 100 1. Shortell SM, Zimmerman JE, Rousseau DM, et al. maximum SOFA score to quantify organ dysfunction/ 90 The performance of intensive care units: does good failure in intensive care: results of a prospective, multi- 80 management make a difference? Med Care. 1994; centre study. Intensive Care Med. 1999;25:686-696. 70 32:508-525. 13. Zweig MH, Campbell G. Receiver-operating char- 60 2. Cullen DJ, Chernow B. Predicting outcome in criti- acteristic (ROC) plots: a fundamental evaluation tool 50 cally ill patients. Crit Care Med. 1994;22:1345- in clinical medicine. Clin Chem. 1993;39:561-577. 40 1348. 14. Bion JF, Aitchison TC, Edlin SA, Ledingham IM.

Mortality, % Mortality, 30 3. Knaus WA, Zimmerman JE, Wagner DP, Draper EA, Sickness scoring and response to treatment as predic- 20 Lawrence DE. APACHE-Acute Physiology and Chronic tors of outcome from critical illness. Intensive Care Med. 10 Health Evaluation: a physiologically based classifica- 1988;14:167-172. tion system. Crit Care Med. 1981;9:591-597. 15. Larvin M, McMahon MJ. APACHE-II score for as- 0 0-1 2-7 8-11 >110-1 2-7 8-11 >11 4. Le Gall JR, Lemeshow S, Saulnier F. A new simpli- sessment and monitoring of acute pancreatitis. Lan- fied acute physiology score (SAPS II) based on a Eu- cet. 1989;2:201-205. Initial SOFA Score ropean/North American multicenter study. JAMA. 16. Sawyer RG, Rosenlof LK, Adams RB, May AK, No. of 1993;270:2957-2963. Spengler MD, Pruett TL. Peritonitis into the 1990s: Patients 5 11 01744 1681 30 5. Lemeshow S, Teres D, Avrunin JS, Gage RW. Re- changing pathogens and changing strategies in the fining intensive care unit outcome prediction by us- critically ill. Am Surg. 1992;58:82-87. ing changing probabilities of mortality. Crit Care Med. 17. Wagner DP, Knaus WA, Harrell FE, Zimmerman 1988;16:470-477. JE, Watts C. Daily prognostic estimates for critically ill 6. Tran DD, Groeneveld ABJ, Vander Meulen J, Nauta adults in intensive care units: results from a prospec- is a good prognostic indicator (espe- JJP, Strack Van Schijndel RJM, Thijs LG. Age, chronic tive multicenter, inception cohort analysis. Crit Care cially the mean and the highest SOFA disease, sepsis, organ system failure, and mortality in Med. 1994;22:1359-1372. a medical intensive care unit. Crit Care Med. 1990; 18. Di Filippo A, De Gaudio AR, Novelli A, et al. Con- scores). Independent of the initial value, 18:474-479. tinuous infusion of vancomycin in methicillin- an increase in the SOFA score during 7. Deitch EA. Multiple organ failure: pathophysiol- resistant staphylococcus infection. Chemotherapy. the first 48 hours of ICU admission pre- ogy and potential future therapy. Ann Surg. 1992; 1998;44:63-68. 216:117-134. 19. Hynninen M, Valtonen M, Markkanen H, et al. dicts a mortality rate of at least 50%. 8. Vincent JL, Moreno R, Takala J, et al, for the Work- Interleukin 1 receptor antagonist and E-selectin con- ing Group on Sepsis-Related Problems of the Euro- centrations: a comparison in patients with severe acute Author Contributions: Study concept and design: pean Society of Intensive Care Medicine. The SOFA pancreatitis and severe sepsis. J Crit Care. 1999;14: Vincent. (Sepsis-related Organ Failure Assessment) score to de- 63-68. Acquisition of data: Ferreira, Bota, Bross. scribe organ dysfunction/failure. Intensive Care Med. 20. Goldhill DR, Sumner A. Outcome of intensive care Analysis and interpretation of data: Ferreira, Bota, 1996;22:707-710. patients in a group of British intensive care units. Crit Me´lot. 9. Regel G, Grotz M, Weltner T, Sturm JA, Tscherne Care Med. 1998;26:1337-1345. Drafting of the manuscript: Ferreira, Vincent. H. Pattern of organ failure following severe trauma. 21. Cryer HG, Leong K, McArthur DL, et al. Multiple Critical revision of the manuscript for important in- World J Surg. 1996;20:422-429. organ failure: by the time you predict it, it’s already tellectual content: Bota. 10. Vincent JL, de Mendonc¸a A, Cantraine F, et al. there. J Trauma. 1999;46:597-604.

Downloaded From: https://jamanetwork.com/ on 09/25/2021