Online Data Supplement:
Mortality Benefit of Recombinant Human Interleukin-1 Receptor Antagonist (rhIL1RA) for Sepsis Varies by Initial IL1RA Plasma Concentration
Supplementary Methods:
Study population – additional detail. Immunosuppression was an exclusion criterion in the
original trial, and was defined as prior solid organ or bone marrow transplant, or significant drug or disease-induced immunosuppression (1). A Clinical Evaluation Committee adjudicated
infectious source and organ dysfunction (1, 2). The primary outcome was 28-day survival and deaths were presumed to be due to sepsis-induced organ failure, however cause of death was not adjudicated at a more granular level. Organ failures were defined as by the parent trial (1, 3) using specific definitions. Septic shock was defined as systolic blood pressure (SBP) of 90 mm Hg or lower, mean arterial pressure of 70 mm Hg or lower, decrease in SBP of 40 mm Hg or greater, or by the use of vasopressor agents at doses exceeding dopamine 5.0 mcg/kg/minute to
maintain blood pressure that persisted despite adequate fluid resuscitation. Disseminated intravascular congestion (DIC) was defined as the presence of both coagulopathy (a
prothrombin time or partial thromboplastin time ≥ 1.2 times normal) and thrombocytopenia, defined as a platelet count equal to or below 100x109/L or a decrease in platelet count ≥ 50%
from baseline. Acute tubular necrosis was defined as serum creatinine ≥ 177 mmol/L (2.0 mg/dL), need for dialysis, or an increase in creatinine ≥ 177 mmol/L (2.0 mg/dL) among subjects with pre-existing renal dysfunction. Acute respiratory distress syndrome (ARDS) was diagnosed using criteria proposed in 1988 (4). Hepatobiliary dysfunction (HBD) was defined as having at least 2 of the following: total serum bilirubin ≥ 43 mmol/L (2.5 mg/dL); serum concentration of
aspartate aminotransferase or alanine aminotransferase ≥ twice the upper limit of normal; or a prothrombin time ≥ 1.5 times the upper limit of normal value. Further, we applied the
macrophage activation syndrome definition applied in a prior analysis of this trial population, which considered subjects adjudicated by the trial’s clinical evaluation committee to have both HBD and DIC (5).
Biomarker assessment: The original trial protocol included a blood draw for exploratory
purposes prior to randomization, and all subjects or their proxies agreed to this aspect as part of their informed consent (1, 6). We had no information on why plasma was available for some subjects but not others, excepting that for approximately 60 subjects, the labeling of the frozen sample had degraded and did not permit an unambiguous match with the clinical database. In an effort to explore whether the subgroup of subjects with plasma were systematically different than subjects who did not have plasma, we undertook 2 supplementary analyses: 1) we tested whether there was a difference in clinical characteristics between subjects with plasma and those without (Supplementary Table E1) and 2) we tested whether the randomization between treatment groups appeared to be preserved among the plasma subgroup (Supplementary Table E2), which we expected to be the case if plasma samples were available at random. Clinical data were absent from the database for another 122 subjects and we had no information on causes for missing data. In addition, we had no ability to compare the clinical characteristics of samples with missing data.
There was an additional confirmatory phase III trial of rhIL1RA in sepsis in 1997 (7), however the study sponsor did not include a similar blood draw for that trial, thus we were unable to test our
Detecting and Reporting Potential Interaction between treatment and plasma biomarker levels:
Our primary analysis was to test for potential heterogeneity in rhIL1RA treatment effect by plasma concentration, which would be indicated by a statistically significant interaction term between the plasma marker and rhIL1RA treatment. We followed a framework recommended by Kent et al to assess and report heterogeneity in treatment effects in clinical trials (8).
Specifically, our a priori statistical plan was to 1) test for statistical interaction between plasma IL1RA or IL1 and rhIL1RA on mortality, and if interaction was present, to 2) perform subgroup analyses stratified by the variables that predicted effect modification. To statistically test for interaction, we considered each plasma marker by decile to simulate a continuous variable and thus maximize the information content of the variable. We regressed observed 28-day mortality including the covariates APACHE III score, rhIL1RA treatment, baseline plasma IL1RA (or IL1) and an interaction term rhIL1RA*IL1RA (IL1). The p-value of this interaction term was our primarily analysis, and we required a stringent alpha level (p < 0.05) to declare significance for the interaction, because there is inadequate justification to liberalize this value even recognizing that few interactions are large enough to detect (9).
To better understand the form of the relationship between mortality, rhIL1RA treatment, and plasma markers and to consider potential categorization of subjects by plasma marker status into marker-positive or marker-negative, we dichotomized the population using the Youden method to maximize the sum of sensitivity and specificity for the plasma marker – mortality relationship. We then tested whether the interaction term remained statistically
significant when treating plasma IL1RA (IL1) as ‘high’ and ‘low’ by this cutpoint in a logistic regression of mortality with the same covariates (APACHE III, rhIL1RA treatment, and the rhIL1RA*plasma IL1RA interaction). To evaluate whether our results depended on our cut point determination, we also tested for statistical interaction between rhIL1RA treatment and plasma IL1RA (IL1) concentration when dichotomizing the population by the median plasma
concentration. To confirm the robustness of the statistical interaction, we also analyzed plasma log(IL1RA) and log(IL1) as continuous variables, and tested models that omitted the APACHE III score.
We next formally tested whether interaction remained present across multiple levels of predicted mortality and presented our results according to baseline risk of mortality (8, 10). This step is strongly recommended in the framework to report potential heterogeneity in treatment effects and it offers information beyond merely adjusting for APACHE III as a covariate, because it tests whether heterogeneity persists despite enriching for high mortality. Adapting the methods of Kent et al, we again performed logistic regression of mortality including covariates rhIL1RA treatment, APACHE III score in tertiles, APACHE III*rhIL1RA treatment, and
rhIL1RA*plasma IL1RA (IL1) category. We report the raw and adjusted mortality results in this stratified analysis. Finally, when interaction remained present, we continued to our secondary analysis, testing by logistic regression whether rhIL1RA associated with mortality in subgroups stratified by the plasma IL1RA (IL1) dichotomization.
Subgroup Analysis: Adjusted Risk Ratios and Adjusted Risk Differences
We used multivariable logistic regression to determine the association of rhIL1RA treatment with mortality, accounting for severity of illness, in strata defined by plasma IL1RA
concentration. Following each stratum-specific logistic regression model, we used post- estimation marginal analysis to convert odds ratios to risk differences by plasma IL1RA concentration (11). Post-estimation marginal analysis uses the estimated model and a set of observations to assign each individual a post-estimation predicted probability of death when the variable of interest (rhIL1RA treatment) equals one, denoted P1, and a predicted probability of death when the variable of interest equals zero (P0). The predicted probabilities for all
subjects are averaged, and the adjusted risk ratio (ARR) is the ratio of the mean predicted probabilities, P1/P0. The adjusted risk difference (ARD) is the difference between the mean predicted probabilities, or P1 - P0 (12). This analysis allows an estimation of the average treatment effect of a single characteristic across all observations while holding other covariates at their original values (13). In our case, we estimated the average treatment effect of rhIL1RA while holding APACHE III score at its original value.
Sensitivity analyses: Plasma IL1 has been described as frequently undetectable during early
sepsis (14), thus we assumed a value of 0.1 pg/ml (roughly 20% of our lowest standard in the ELISA assay) for any samples with a value that was out of range low. We then repeated analyses excluding these subjects for all IL1 analyses. We also performed sensitivity analyses excluding 26 subjects who had both disseminated intravascular coagulopathy and hepatobiliary
dysfunction because these patients might have macrophage activation syndrome (MAS). A prior publication has demonstrated a favorable response to rhIL1RA among these subjects (5), and we wished to ensure that our detection of a statistical interaction was not due to MAS.
Exploratory analyses – combinatorial variables of IL1RA and IL1: To test whether a ratio of
IL1RA/IL1 or a product of IL1RA*IL1 functioned as markers of plasma IL-1 pathway activation to enrich for rhIL1RA response, we derived each of these ratios and tested for statistical
interaction with rhIL1RA treatment on the mortality effect as described above for plasma IL1RA and plasma IL1. Each variable was tested for interaction as deciles, dichotomized at the median value observed, or (for IL1RA*IL1 only given its association with mortality), at the Youden-defined cut point that optimized a mortality AUROC. A p-value of 0.05 was considered significant.
Supplemental Results
Table E1 Depicts the rhIL1RA patient population stratified by the availability of plasma.
Continuous data are compared by the Wilcoxon rank sum test and categorical data by the chi square test as appropriate. No significant differences were noted between subjects with and without plasma.
Table E2 evaluates baseline characteristics among the plasma-evaluable group with respect to treatment allocation, to estimate whether randomization was imbalanced among the subjects in whom plasma IL1RA and IL1 was measured.
Table E3 presents the population stratified by the Youden-determined cut point in plasma IL1RA concentration that optimizes the receiver operating characteristics of plasma IL1RA for mortality. For 54% of the population, baseline plasma IL1RA concentration was above
2071 pg/ml, and this group had significantly higher rates of organ failure and mortality. In addition, plasma IL1 concentration was significantly higher in the group defined by plasma IL1RA > 2071 pg/ml. If the population was dichotomized at the median plasma IL1RA
concentration of 2455 pg/ml, rather than the data-derived cut point, the high IL1RA group had a lower mortality with rhIL1RA treatment as shown in Table E4, whereas the low plasma IL1RA group had slightly higher mortality with rhIL1RA, though this latter effect was not statistically significant.
Because the original trial involved randomization to one of 2 rhIL1RA doses, we tested whether there was evidence for a dose-response relationship in the interaction between rhIL1RA and plasma IL1RA concentration on mortality. As shown in Tables E5a and E5b, the magnitude of interaction effect appeared very similar in stratified analysis of each rhIL1RA dose, thus we combined both doses into a single “rhIL1RA” group.
Table E6 provides the raw mortality data and APACHE III-adjusted risk difference (ARD) and adjusted relative risk of rhIL1RA treatment on mortality when the population was stratified by Youden-determined plasma IL1 cut point, 8 pg/ml. Although the direction of benefit was favorable for the group defined by high plasma IL1, this effect was not statistically significant.
Results were unchanged when the population was dichotomized at the median IL1 level, with the high plasma IL1 group has an adjusted risk difference for mortality of -0.17 (95% CI -0.13, 0.10), p=0.77.
Table E7 displays the p-values for interaction for derived combinatorial variables IL1RA/IL1 and IL1RA*IL1. For each derived variable, we did not detect any significant interaction with rhIL1RA treatment on sepsis mortality.
Table E1: Patient characteristics and outcomes for subjects based on plasma availability. The subpopulation with plasma available did not differ significantly from those without plasma with the exception of a lower ARDS incidence for the plasma subset. Organ dysfunctions are defined in the text of the online supplement (page 1-2) as defined by the parent trials (1, 5).
Plasma measured (n=529)
No Plasma Available
(n=242) p-value
Treatment allocation Placebo
rhIL1RA 0.1 mg/kg/hr rhIL1RA 0.2 mg/kg/hr
164 183 182
93 77 72
0.121
Age 63 (45 – 71) 62 (45 – 70) 0.614
Female Gender 318 (60%) 138 (57%) 0.418
APACHE III predicted
mortality risk 0.32 (0.19, 0.48) 0.32 (0.21, 0.47) 0.915
Infection source:
Gram negative Gram positive Mixed bacterial Other
Unknown
136 (26%) 100 (19%) 156 (29%) 35 (7%) 102 (19%)
66 (27%) 40 (17%) 79 (33%) 9 (4%) 48 (20%)
0.451
Septic Shock 428 (81%) 192 (80%) 0.687
Acute respiratory
distress syndrome 127 (24%) 72 (30%) 0.096
Disseminated intravascular coagulopathy
69 (13%) 37 (15%) 0.401
Biliary dysfunction 132 (25%) 72 (30%) 0.161
Macrophage activation
syndrome 26 (5%) 17 (7%) 0.236
Acute tubular necrosis 166 (31%) 68 (28%) 0.358
Mortality at 7 days 82 (16%) 36 (15%) 0.823
Mortality at 28 days 167 (32%) 72 (30%) 0.613
Table E2: Patient characteristics of the plasma-evaluable group with respect to treatment randomization. Continuous data were compared by the Wilcoxon rank sum test and categorical data by chi square test. We observed no imbalance in age, gender, APACHE III-predicted
mortality risk, or infectious etiology between treatment groups.
Placebo (n=164)
rhIL1RA
(n=365) p-value
Age 64 (48 – 71) 62 (43 – 71) 0.336
Female Gender
n, (%) 67 (41%) 144 (39%) 0.761
APACHE III predicted
mortality risk 0.28 (0.19, 0.47) 0.32 (0.20, 0.49) 0.284
Infection source:
Gram negative Gram positive Mixed bacterial Other
Unknown
41 (25%) 27 (16%) 48 (29%) 14 (9%) 34 (20%)
95 (26%) 73 (20%) 108 (30%)
21 (6%) 68 (19%)
0.342
Septic Shock 135 (82%) 293 (80%) 0.580
Mortality day 28 54 (32.9%) 113 (31.0%) 0.652
Table E3: Patients with elevated baseline plasma IL1RA had a higher severity of illness and more organ dysfunction. Population dichotomized by the Youden-derived cut point for IL1RA, 2071 pg/ml. Continuous data were compared by Wilcoxon rank sum test and categorical data by chi square test.
High plasma IL1RA (n = 286)
Low plasma IL1RA
(n = 243) p-value
APACHE III-Predicted mortality
0.36 (0.22 – 0.56)
0.27
(0.17 – 0.42) < 0.001
Septic Shock 251 (88%) 177 (73%) < 0.001
Acute respiratory
distress syndrome 58 (20%) 69 (29%) 0.029
Disseminated intravascular coagulopathy
44 (15%) 25 (10%) 0.083
Biliary Dysfunction 83 (29%) 49 (20%) 0.019
Macrophage
activation syndrome 19 (7%) 7 (3%) 0.046
Acute tubular necrosis 115 (41%) 50 (21%) < 0.001
IL1 (pg/ml) 7.6 (1.0 – 32.0) 0.78 (0.1 – 5.7) < 0.001
IL1RA / IL1 Ratio 1990
(279 - 13097)
372
(27 - 1666) < 0.001
IL1RA*IL1 80511
(6373 - 630352)
177
(16 - 2956) < 0.001
Table E4: Effect modification of rhIL1RA treatment effect by plasma IL1RA concentration is still present when the population is dichotomized by the median plasma IL1RA level. The population was stratified at plasma IL1RA concentration = 2454 pg/ml. The p-value of the interaction term rhIL1RA treatment*plasma IL1RA_median was significant (p=0.036), thus we performed stratified analyses. Using logistic regression accounting for APACHE III score and rhIL1RA treatment, and post-estimation analysis to convert odds ratios to adjusted relative risks and adjusted risk differences, we analyzed strata defined by plasma IL1RA concentration.
Plasma IL1RA
by Cut point Mortality Rate
Adjusted Risk Difference
(95% CI)
Adjusted Relative Risk of Mortality
(95% CI)
p-value Placebo rhIL1RA
Low (n=266)
< 2454 pg/ml
18 / 82 (22%)
53 / 184
(29%) +0.05 (-0.05, 0.16) 1.23 (0.80, 1.91) p=0.327 High (n=263)
≥ 2454 pg/ml
36 / 82 (44%)
60 / 181
(33%) -0.12 (-0.24, -0.01) 0.73 (0.54, 0.98) p=0.035
Tables E5a and E5b: Lack of evidence for a dose response effect in the interaction between rhIL1RA and baseline plasma IL1RA level. We display the stratified analysis for placebo group compared to rhIL1RA dose 1.0 mg/kg/hr group, in Table E4a, and for placebo compared to rhIL1RA 2.0 mg/kg/hr, in Table E4b. Thus, we combined both doses of rhIL1RA for interaction analysis.
Table E5a: Significant effect modification of rhIL1RA treatment effect by baseline plasma IL1RA concentration comparing low-dose (1 mg/kg/hr) rhIL1RA to placebo.
Mantel-Haenszel test for inhomogeneity of odds ratios: p=0.067.
Plasma IL1RA
by Cut point Mortality Rate OR Mortality
(95% CI) Placebo rhIL1RA
Low (n=156)
< 2017 pg/ml
14 / 76 (18%)
23 / 80
(29%) 1.50 (0.70 – 3.20)
High (n=191)
≥ 2017 pg/ml
40 / 88 (45%)
36 / 103
(35%) 0.61 (0.33 – 1.11)
Table E5b: Significant effect modification of rhIL1RA treatment effect by baseline plasma IL1RA concentration comparing high-dose (2 mg/kg/hr) rhIL1RA to placebo.
Mantel-Haenszel test for inhomogeneity of odds ratios: p=0.036.
Plasma IL1RA
by Cut point Mortality Rate OR Mortality
(95% CI) Placebo rhIL1RA
Low (n=163)
< 2017 pg/ml
14 / 76 (18%)
22 / 87
(25%) 1.78 (0.83 – 3.83)
High (n=183)
≥ 2017 pg/ml
40 / 88 (45%)
32 / 95
(34%) 0.64 (0.36 – 1.16)
Table E6: Treatment effect of rhIL1RA stratified by baseline plasma IL1 concentration,
assuming a value of 0.10 pg/ml for 125 subjects with undetectable IL1 level. The interaction term [rhIL1RA treatment*IL1 concentration] was not statistically significant: p=0.26. The population was dichotomized at the empiric cut point that optimized the area under the receiver operating curve for mortality: plasma IL1= 8 pg/ml. Plasma IL1 was above this threshold for 38% of subjects, and these subjects may have benefited from rhIL1RA. In a
sensitivity analysis, subjects with undetectable plasma IL1 were excluded, which decreased the low plasma IL1group to 206 subjects with an adjusted relative risk of 0.98 (95% CI 0.64 to 1.50), p=0.98.
Plasma IL1
by Cut point Mortality Rate
Adjusted Risk Difference
(95% CI)
Adjusted Relative Risk of Mortality
(95% CI)
p-value Placebo rhIL1RA
Low (n=331) 27 / 99 (27%)
68 / 232
(29%) +0.01 (-0.09, 0.11) 1.02 (0.72, 1.45) p=0.91 High (n=198) 27 / 64
(42%)
44 / 134
(33%) -0.10 (-0.24, 0.03) 0.76 (0.54, 1.07) p=0.14
Table 7: Derived combinatorial variables IL1RA/IL1 or IL1RA*IL1 fail to display significant interaction with rhIL1RA treatment. The logistic regression model included the interaction term of each derived variable with rhIL1RA and the APACHE III score. Results were unchanged with the omission of APACHE III score.
Variable considered for interaction testing Interaction term p-value
IL1RA/IL1 Decile 0.924
Median value 0.610
IL1RA*IL1 Decile 0.061
Median value 0.292
Youden cut point for mortality 0.246
SUPPLEMENTAL REFERENCES
1. Fisher, C.J., Jr., Dhainaut, J.F., Opal, S.M., Pribble, J.P., Balk, R.A., Slotman, G.J., Iberti, T.J., Rackow, E.C., Shapiro, M.J., Greenman, R.L., et al. 1994. Recombinant human interleukin 1 receptor antagonist in the treatment of patients with sepsis syndrome. Results from a randomized, double-blind, placebo-controlled trial. Phase III rhIL-1ra Sepsis Syndrome Study Group. JAMA 271:1836-1843.
2. Knaus, W.A., Harrell, F.E., Fisher, C.J., Jr., Wagner, D.P., Opal, S.M., Sadoff, J.C., Draper, E.A., Walawander, C.A., Conboy, K., and Grasela, T.H. 1993. The clinical evaluation of new drugs for sepsis. A prospective study design based on survival analysis. JAMA 270:1233-1241.
3. Fisher, C.J., Jr., Slotman, G.J., Opal, S.M., Pribble, J.P., Bone, R.C., Emmanuel, G., Ng, D.,
Bloedow, D.C., and Catalano, M.A. 1994. Initial evaluation of human recombinant interleukin-1 receptor antagonist in the treatment of sepsis syndrome: a randomized, open-label, placebo- controlled multicenter trial. Crit Care Med 22:12-21.
4. Murray, J.F., Matthay, M.A., Luce, J.M., and Flick, M.R. 1988. An Expanded Definition of the Adult Respiratory Distress Syndrome. American Review of Respiratory Disease 138:720-723.
5. Shakoory, B., Carcillo, J.A., Chatham, W.W., Amdur, R.L., Zhao, H., Dinarello, C.A., Cron, R.Q., and Opal, S.M. 2016. Interleukin-1 Receptor Blockade Is Associated With Reduced Mortality in Sepsis Patients With Features of Macrophage Activation Syndrome: Reanalysis of a Prior Phase III Trial*. Critical Care Medicine 44:275-281.
6. Opal, S.M., Scannon, P.J., Vincent, J.-L., White, M., Carroll, S.F., Palardy, J.E., Parejo, N.A., Pribble, J.P., and Lemke, J.H. 1999. Relationship between Plasma Levels of Lipopolysaccharide (LPS) and LPS-Binding Protein in Patients with Severe Sepsis and Septic Shock. Journal of Infectious Diseases 180:1584-1589.
7. Opal, S.M., Fisher, C.J., Dhainaut, J.F., Vincent, J.L., Brase, R., Lowry, S.F., Sadoff, J.C., Slotman, G.J., Levy, H., Balk, R.A., et al. 1997. Confirmatory interleukin-1 receptor antagonist trial in severe sepsis: A phase III, randomized, doubleblind, placebo-controlled, multicenter trial.
Critical Care Medicine 25:1115-1124.
8. Kent, D.M., Rothwell, P.M., Ioannidis, J.P., Altman, D.G., and Hayward, R.A. 2010. Assessing and reporting heterogeneity in treatment effects in clinical trials: a proposal. Trials 11:1-11.
9. Brookes, S.T., Whitely, E., Egger, M., Smith, G.D., Mulheran, P.A., and Peters, T.J. 2004.
Subgroup analyses in randomized trials: risks of subgroup-specific analyses; power and sample size for the interaction test. J Clin Epidemiol 57.
10. Kent, D.M., Alsheikh-Ali, A.A., and Hayward, R.A. 2008. Competing risk and heterogeneity of treatment effect in clinical trials. Trials 9.
11. Graubard, B.I., and Korn, E.L. 1999. Predictive margins with survey data. Biometrics 55:652- 659.
12. Norton, E.C., Miller, M.M., and Kleinman, L.C. 2013. Computing adjusted risk ratios and risk differences in Stata. Stata Journal 13:492-509.
13. Shashaty, M.G.S., Meyer, N.J., Localio, A.R., Gallop, R., Bellamy, S.L., Holena, D.N., Lanken, P.N., Kaplan, S., Yarar, D., Kawut, S.M., et al. 2012. African American race, obesity, and blood product transfusion are risk factors for acute kidney injury in critically ill trauma patients.
Journal of Critical Care 27:496-504.
14. Goldie, A.S., Fearon, K.C., Ross, J.A., Barclay, G.R., Jackson, R.E., Grant, I.S., Ramsay, G., Blyth, A.S., and Howie, J.C. 1995. Natural cytokine antagonists and endogenous antiendotoxin core antibodies in sepsis syndrome. The Sepsis Intervention Group. JAMA 274:172-177