THERAPEUTIC HYPOTHERMIA AND TEMPERATURE MANAGEMENT Volume 2, Number 1, 2012 ª Mary Ann Liebert, Inc. DOI: 10.1089/ther.2012.0003
Evaluating Traditional Prognostic Measures in Patients Undergoing Hypothermia After Cardiac Arrest John C. O’Horo, M.D.,1 Mihail Andreev, M.D.,2 Wael Hassan, M.D.,1 and Asif Anwar, M.D., M.S., FCCP 3*
Background: Therapeutic hypothermia is one of the few interventions that improve mortality and neurologic outcomes in patients who have experienced cardiac arrest. There is a lack of validated tools to predict survival in patients who have undergone hypothermia after cardiac arrest (HACA). Methods: A retrospective analysis was performed of all patients who underwent HACA at Aurora St. Luke’s Medical Center (ASLMC) since the protocol was implemented in September 2008. Initial rhythm, whether percutaneous coronary intervention (PCI) was performed, lactate levels, duration of resuscitation, and APACHE II scores were compared for survivors and non-survivors, and a logit binary regression model was constructed. Results: A total of 143 patients were identified and had data abstracted. APACHE-II, duration of resuscitation, and initial rhythm were all strongly correlated with survival. Initial serum lactate levels were higher in nonsurvivors than survivors ( p = 0.005), though the trend of lactate change at next draw was not predictive. Quantitative TnI was not significantly different between arms. Conclusion: Lactate levels show promise as a biomarker for survival in HACA patients resuscitation length, presence of PCI, and APACHE-II scores can provide good prognostic information, even in the early hours following a resuscitation event.
demonstrated promise in the HACA population in three studies that drew lactate levels post-resuscitation. The first was an observational study, which noted a correlation between lactate levels and survival (Oddo et al., 2008). The second was a prospective study, which found that duration of resuscitation and lactate levels were the only factors significantly predictive of survival (Shinozaki et al., 2011). The final study, a multicenter retrospective study, measured lactate 1 hour after completing resuscitation, and found a stepwise association with lactate levels and mortality, with lactates > 10 mmol/L associated with 92% mortality (Cocchi et al., 2001). The HACA protocol currently used at Aurora St. Luke’s Medical Center (ASLMC), a 730-bed tertiary referral hospital in Milwaukee, has routinely determined lactate, troponin I (TnI), and APACHE-II data on all patients since approval of the local protocol in September 2008. This study retrospectively analyzes these data from this patient population.
Introduction
M
ore than 300,000 cardiac arrests occur each year in the United States, with a dismal overall survival rate of less than 10% (Nichol et al., 2008). Hypothermia after cardiac arrest (HACA) is one of the few interventions that has been convincingly demonstrated to improve mortality and neurologic outcomes in patients who have experienced cardiac arrest (Bernard et al., 2002; HACA Study Group, 2002). In spite of this, mortality remains high, and it remains difficult to predict which patients will benefit from the procedure. Critical care physicians are asked to provide a prognosis for HACA patients. Unfortunately, this is quite difficult due to a lack of validated tools. Traditional neurologic assessments do not always provide adequate prognostic information (Young, 2009; Bisschops et al., 2011), and studies into traditional biomarkers and scoring tools have largely not been tested in the HACA population. Acute physiology and chronic health evaluation II (APACHE-II) score and lactate levels have both been shown to be predictive in the general cardiac arrest population (Schuetz et al., 2010; Cocchi, 2011) and lactate levels specifically have
Methods ASLMC implemented a HACA protocol in September 2008 (see Fig. 1). This protocol underwent no changes during the
1
Aurora UW Medical Group, Milwaukee, Wisconsin. Aurora Medical Group, Department of Hospital Medicine, Milwaukee, Wisconsin. 3 Aurora Medical Group, Department of Critical Care, Milwaukee, Wisconsin. *Current additional affiliations: University of Wisconsin School of Medicine and Public Health, Madison, Wisconsin; United Health Systems, Kenosha, Wisconsin. 2
24
PROGNOSIS TOOLS IN HYPOTHERMIA AFTER CARDIAC ARREST
25
Statistical analysis was performed using Minitab 15 software. Correlation between survivorship was independently tested for each variable using one-way analysis of variance (ANOVA) testing. Several logit binary regression models were constructed from these data for comparison of the relative importance of each variable. Results
FIG. 1. Interval plot of survival by year. No statistically significant difference was seen ( p = 0.34). Mean survival followed the linear regression 175.47–0.0767(x) where x was the year of the code (R2 = 0.96).
One hundred and forty-four patients were started on the HACA protocol at ASLMC between September 2008 and April 2011 and included in the study. Of these, 91 (63.2%) were categorized as non-survivors, with 84 (58.3%) patients expiring and seven (5%) being transferred to a hospice. The study identified 53 (37.7%) survivors with 29 (20%) being discharged home, five (3.5%) to a long-term care facility, 15 (10%) to rehab, two (1.4%) to a skilled nursing facility, and one (0.7%) leaving against medical advice. Trend data
study period. Since the initiation of this protocol, patients undergoing therapeutic hypothermia have been tracked in a database. Following institutional review board approval, patient names and admissions were obtained from this database, and accessed in the electronic medical record system. Data were abstracted by three of the researchers independently. All information was abstracted twice, and conflicts in the resultant data were resolved with discussion among the researchers. Subjects were first categorized by their discharge disposition, with those who expired in-house or transferred to hospice being categorized as non-survivors, and the rest as survivors. On each patient, the rhythm at initiation of resuscitative efforts and duration of resuscitation was abstracted from a quality database maintained for HACA patients. Rhythms were coded as able to be defibrillated (ventricular tachycardia and fibrillation) and unable to defibrillate (asystole and pulseless electrical activity). Patients who were shocked with an automatic external defibrillator in the field without presenting rhythm were coded as able to defibrillate. Data on time down before resuscitation were initiated on out-of-hospital arrests were not available. Other data were abstracted from the electronic medical record system. Procedure notes were reviewed on all patients who underwent coronary catheterization. Those who underwent percutaneous coronary intervention (PCI) were noted as such in the data abstraction. First lactate level drawn by the lab was abstracted for each patient who had one drawn within 24 hours of code event. Point-of-care and bedside testing values were excluded because of a high variability seen in these, and questions about their quality at this institution. When possible, second lactate and the rate of change of lactate were also abstracted. First quantitative TnI were also abstracted. Originally, procalcitonin levels were set to be abstracted as well, but no HACA patients had this level checked. Finally, APACHE-II scores were calculated for each patient and determinants abstracted from the 24 hours of data preceding cooling. For patients in hospital less than 24 hours before the cooling was started, the most abnormal set of labs and vitals between admission and cooling was abstracted. The final APACHE-II score was calculated using the tool at www.sfar.org/scores2/apache22.html.
HACA patients were stratified by year to analyze if there was a significant difference in survival over the study period. In 2008, 11 patients were started on HACA and 45.5% survived. In 2009, 53 HACA protocols were initiated and 41.5% survived. The year 2010 saw 61 HACA patients with a 31.2% survival rate. Although a linear regression model demonstrated an 8% drop in survival each year (R2 = 0.96), this trend was not significant by ANOVA ( p = 0.34). This is illustrated in Figure 1. Initial rhythm and PCI Initial rhythm was significantly different between survivors and non-survivors. Asystole had the lowest survival and ventricular fibrillation and tachycardia were tied for best survival ( p < 0.001). When coded as shockable versus not, shockable rhythms had significantly better survival than nonshockable rhythms ( p < 0.001). This is depicted in Figure 2. Thirty-six patients underwent PCI after their cardiac arrest, with a survival rate of 55.6% compared with 29.2% in those who did not have percutaneous coronary intervention ( p = 0.004).
FIG. 2. Survival rates by rhythm. Better survival was seen in rhythms able to be defibrillated. VF and VT patients had a combined 52.3% survival rate compared to 10.7% in combined PEA/asystole patients ( p < 0.001).
26
O’HORO ET AL.
Lactate levels and rate of change Eighty-eight (62%) subjects had a first lactate level drawn. Time to first draw after admission was not different between survivors and non-survivors ( p = 0.07). First lactate level differed significantly ( p = 0.005) between arms, with a mean level of 3.4 mmol/L (SD = 2.1 mmol/L) in survivors and 5.9 mmol/ L (SD = 4.5 mmol/L) in non-survivors. Second lactate levels were drawn on 49 (34%) patients, also without significantly different times to draw ( p = 0.752). Again, results failed to achieve statistical significance, with a mean level of 3.4 mmol/L (SD = 1.9 mmol/L) in survivors, and 4.1 mmol/L (SD = 3.8 mmol/L) in non-survivors ( p = 0.53). Rates of lactate change were calculated for all 49 patients with a second lactate. Results again failed to achieve significance with a mean change of - 0.15 mmol/L per hour (SD = 0.37 mmol/L per hour) in survivors and - 0.27 mmol/L per hour (SD = 0.76 mmol/L per hour) in non-survivors ( p = 0.59). Quantitative TnI Quantitative TnI level was determined in 125 patients. Levels were not significantly different between groups ( p = 0.481); survivors averaged 17.37 ng/mL (SD = 55.44 ng/ mL) while non-survivors averaged 26.01 ng/mL (SD = 55.33 ng/mL). The validity of this observation is questionable in light of significantly different draw times, with survivors having TnI levels drawn 7.4 hours after code versus nonsurvivors 3.6 hours after code ( p = 0.03). Duration of resuscitation Duration of resuscitation was available on code documentation for 121 patients. Survivors had significantly shorter resuscitation times, averaging 18.3 (SD = 20.1) minutes versus 34.0 (SD = 27.6) minutes in non-survivors ( p = 0.001). APACHE II data Individual factors in the APACHE-II score are detailed in Table 1. A strong correlation was noted with mean arterial pressure, pH, creatinine, hematocrit, and Glasgow Coma Scale, though the aggregated APACHE-II score was the most predictive by far. Paired t-tests and Pearson correlation coef-
ficients demonstrate that the APACHE-II predicted death rate is similar to what is observed in this study (see Fig. 2). Binary regression modeling Several regression models were developed using combinations of statistically significant variables (see Table 2). These results were consistent with the ANOVA analysis. Length of resuscitation, lactate levels, and APACHE II score were negatively associated with survival, while rhythms subject to defibrillation and undergoing PCI were associated with decreased mortality. The relative contributions to each of these under the PCI/ duration of resuscitation/first lactate/APACHE II/Initial Rhythm model is modeled in Figure 2. This model demonstrated a good correlation with observed variables, with a Sommer’s D statistic of 0.85 and a Hosmer–Lemeshow statistic of 0.775, failing to reject the hypothesis that the model does not fit the data. All attempts at modeling component variables of the APACHE II score demonstrated inferior predictive value relative to the unaltered APACHE II score (see Fig. 3). Likewise, models including the TnI, the rate of lactate change, and second lactate all decreased the correlation with observed data. Marginal effects were tested with the model, demonstrating linear relationships with the variables in the model. The most significant relative variable was duration of resuscitation, followed by APACHE score with a relatively minor contribution of lactate levels. The categorical variables, PCI, and ability to defibrillate had the most impact on the predicted outcome. Discussion Although subject to limitations inherent to its retrospective design, this study identifies several useful predictors for survival in the minutes following the decision to initiate hypothermia protocol. In addition, this also illustrates the need for caution when interpreting the prognostic value of lactate for the HACA population. Although lactate levels did achieve statistical significance in this and other studies (Young, 2009; Bisschops et al., 2011), the logistic regression models demonstrate that lactate is a relatively minor contributor compared to APACHE II levels,
Table 1. Characteristics of Survivors Versus Non-survivors Variable Age (years) APACHE-II score Lactate (mmol/L) Proportion in-house arrest Proportion male Duration of resuscitation (min)
Mean 63.4 60.4 29.8 25.8 6.92 5.51 0.10 0.13 0.63 0.78 34.0 18.4
(non-survivors) (survivors) (non-survivors) (survivors) (non-survivors) (survivors) (non-survivors) (survivors) (non-survivors) (survivors) (non-survivors) (survivors)
SD
p value
14.8 16.7 6.5 6.7 4.32 3.26 0.31 0.34 0.05
0.264
27.6 20.6
0.002
0.001 0.092 0.625 0.08
P values are calculated using a one-way ANOVA model. APACHE-II score was calculated using the Socie´te´ Franc¸aise d’Anesthe´sie de Re´animation APACHE II calculator at www.sfar.org/scores2/apache22.html. SD, standard deviation.
PROGNOSIS TOOLS IN HYPOTHERMIA AFTER CARDIAC ARREST
27
Table 2. Logistic Regression Models Model Lactate predictive model
Lactate/TnI/APACHE II model
Duration of resuscitation/ rhythm/PCI model
APACHE constituent model
Restricted APACHE model
Restricted APACHE score contribution model
Duration of resuscitation/ Able to defibrillate/PCI model PCI/ Duration of resuscitation/ Lactate/ APACHE II/ Initial Rhythm
PCI/
Logistic regression model variables included Constant First lactate Next lactate Rate of lactate change First lactate2 Constant First lactate TnI APACHE-II score Constant Initial rhythm PEA VT VF PCI Duration of resuscitation Constant Temperature MAP HR RR FiO2 PaO2 pH Na K Creatinine ARF present HCT WBC GCS Age Chronic disease Constant MAP pH Creatinine Hct Constant MAP APACHE contribution pH APACHE contribution Hematocrit APACHE contribution Constant PCI Shockable rhythm Duration of resuscitation Constant Duration of resuscitation PCI Lactate APACHE-II Initial Rhythm PEA VF VT Constant
Variable p values
Hosmer–Lemeshow goodness of fit test
Sommer’s D statistic
- 7.26 - 0.025 - 0.067 0.261 0.0003 1.047 - 0.224 - 0.005 - 0.020 - 2.058 1.969 2.969 3.781 1.613 - 0.085
0.136 0.920 0.759 0.826 0.158 0.374 0.031 0.431 0.634 0.073 0.134 0.001 0.023 0.016 0.001
0.444
0.28
0.189
0.39
0.691
0.79
- 34.1 0.043 0.009 0.010 - 0.021 - 0.004 - 0.002 2.92 0.047 0.079 - 0.665 - 0.21 0.031 0.013 0.143 - 0.001 - 0.031 - 18.9 0.010 2.34 - 0.660 0.438 0.080 0.064 - 0.232 - 0.373 - 0.948 1.372 2.564 - 0.0868 - 19.14 - 0.068 1.177 - 0.024 - 0.032
0.019 0.547 0.286 0.268 0.579 0.789 0.293 0.044 0.242 0.736 0.061 0.608 0.288 0.699 0.237 0.967 0.960 0.037 0.222 0.062 0.033 0.107 0.808 0.651 0.046 0.109 0.146 0.019 0.000 0.000 0.997 0.018 0.280 0.854 0.592
0.483
0.51
0.592
0.42
0.669
0.31
0.500
0.78
0.775
0.85
19.98 22.09 20.80 - 0.677
0.997 0.997 0.997 0.699
0.069
0.81
Coefficient
(continued)
28
O’HORO ET AL. Table 2. (Continued) Logistic regression model variables included
Model Duration of resuscitation/ Lactate/ APACHE II/ Able to defibrillate
PCI Duration of resuscitation Lactate APACHE II Able to Defibrillate
Coefficient
Variable p values
1.060
0.274
- 0.063 - 0.142 - 0.003 2.988
0.020 0.237 0.000
Hosmer–Lemeshow goodness of fit test
Sommer’s D statistic
Coefficients for logit function, p values, and relative goodness of fit for all models tested shown. PCI, percutaneous coronary intervention.
FIG. 3. Dotplot of APACHE II score for survivors versus non-survivors. Paired t-test comparing to APACHE-II predicted death rate shows a high degree of similarity ( p = 0.631). Pearson correlation coefficient = 0.777 ( p = 0.0000).
duration of resuscitation, and ability to defibrillate rhythm. Lactate levels, though useful, should only be interpreted within the context of these other factors. Duration of resuscitation was highly predictive of survival, with longer times to spontaneous return to circulation being strongly correlated with higher mortality. This is consistent with other studies into resuscitation (Rosenberg et al., 1993). The utility of the APACHE II score was remarkable. Despite several attempts at modeling, no combination of the component variables was more predictive than the APACHE score itself. Previous data have demonstrated their predictive value in post-resuscitation patients (Bialecki and Woodward, 1995), but it has never been validated in the HACA population specifically before. A previous study has found the degree of quantitative TnI elevation to correlate with survival ( Joly et al., 2011). This was not replicated in this study. Several factors may have contributed to this. The TnI was drawn significantly earlier, on average, in patients who did not survive. Given the rapid peaking and declining of TnI, this severely limits the validity of comparing these data. Another factor is the hospital itself; ASLMC is a cardiology referral center with a 24-hour cardiac catheterization lab. Many of the STEMI related cardiac arrest patients were brought to ASLMC to receive coronary intervention. These patients would tend to have higher TnI, and the predictive effect of this would be blunted by the benefit of PCI. Several other biomarkers have been proposed as tools in predicting survival in HACA patients, including procalcitonin, cell-free plasma DNA levels (Amalich et al., 2010), neuron specific enolase, and S-100b (Sanfilippo et al., 2010). None of the patients in the study period had any of these values determined. Future studies to assess these potential prognostic tools are needed. The overall survival rate seen in HACA patients at ASLMC was 37.7%. This is significantly lower than reported mean of 65% seen in a review article (Sagalyn et al., 2009). The exact reason for this is unclear. Future multicenter trials would
better be able to determine the relative predictive value of these factors in HACA patients. Despite the limits inherent to a retrospective study, these data can be of value to clinicians. When providing early prognostic information for HACA patients, APACHE-II criteria and length of resuscitation efforts can provide useful tools. Lactate levels, while useful, should be interpreted with caution, pending further prospective study into this additional other biomarkers. Acknowledgments The authors would like to thank Sara Marzinski, RN, for providing access to the hypothermia protocol patient database. Disclosure Statement No competing financial interests exist. References Amalich F, Mene´ndez M, Lagos V, Ciria E, Quesada A, Codoceo R, Vazquez JJ, Lo´pez-Collazo E, Montiel C, et al. Prognostic value of cell-free plasma DNA in patients with cardiac arrest outside the hospital: an observational cohort study. Crit Care 2010:14:R47. Bernard SA, Gray TW, Buist MD, Jones BM, Silvester W, Gutteridge G, et al. Treatment of comatose survivors of out-ofhospital cardiac arrest with induced hypothermia. N Engl J Med 2002;346:557–563. Bialecki L, Woodward RS. Predicting death after CPR. Experience at a nonteaching community hospital with a full time critical care staff. Chest 1995;108:1009–1117. Bisschops LL, van Alfen N, Bons S, van der Hoeven JG, Hoedemaekers CW. Predictors of poor neurologic outcome in patients after cardiac arrest treated with hypothermia: a retrospective study. Resuscitation. 2011;82:696–701. Cocchi MN, Miller J, Hunziker S, et al. The association of lactate and vasopressor need for mortality prediction in survivors of cardiac arrest. Minerva Anestesiol 2001;77:1063–1071.
PROGNOSIS TOOLS IN HYPOTHERMIA AFTER CARDIAC ARREST Cocchi MN, Miller J, Hunziker S, Carney E, Salciccioli J, Farris S, et al. The association of lactate and vasopressor need for mortality prediction in survivors of cardiac arrest. Minerva Anestesiol 2011;11:1063–1071. Hypothermia After Cardiac Arrest Study Group. Mild therapeutic hypothermia to improve the neurologic outcome after cardiac arrest. New Engl J Med 2002;346:550–556. Joly SS, Shenkman H, Brieger D, Fox KA, Yan AT, Eagle KA, Steg PG, Lim KD, Quill A, Goodman SG, GRACE investigators. Quantitative troponin and death, cardiogenic shock cardiac arrest and new heart failure in patients with non-STsegment elevation acute coronary syndromes (NSTE ACS): insights from the Global Registry of Acute Coronary Events. Heart 2011;97:197–202. Nichol G, Thomas E, Callaway CW, Hedges J, Powell JL, Aufderheide TP, et al. Regional variation in out-of-hospital cardiac arrest incidence and outcome. JAMA 2008;300:1423–1431. Oddo M, Ribordy V, Feihl F, Rossetti AO, Schaller MD, Chiolero R, et al. Early predictors of outcome in comatose survivors of ventricular fibrillation and non-ventricular fibrillation cardiac arrest treated with hypothermia: A prospective study. Crit Care Med 2008;36:2296–2301. Rosenberg M, Wang C, Hoffman-Wilde S, Hickam, D, et al. Results of cardiopulmonary resuscitation. Failure to predict survival in two community hospitals. Arch Intern Med 1993;154:1370–1375.
29
Sagalyn E, Band RA, Gaieski DF, et al. Therapeutic hypothermia after cardiac arrest in clinical practice: review and compliation of recent experiences. Crit Care Med 2009;37:223–226. Sanfilippo F, Li Volti G, Ristagno G, Murabito P, Pellis T, Astuto M, Gullo A, et al. Clinical biomarkers in brain injury: a lesson from cardiac arrest. Front Biosci (Schol Ed) 2010;2623–2640. Schuetz P, Affolter B, Hunziker S, Winterhalder C, Fischer M, Balestra GM, et al. Serum procalcitonin, C-reactive protein and white blood cell levels following hypothermia after cardiac arrest: a retrospective cohort study. Eur J Clin Invest 2010;40:376–381. Shinozaki K, Oda S, Sadahiro T, Nakamura M, Hirayama Y, Watanabe E, et al. Blood ammonia and lactate levels on hospital arrival as a predictive biomarker in patients with out-ofhospital cardiac arrest. Resuscitation 2011;82:404–409. Young GB. Clinical practice. Neurologic prognosis after cardiac arrest. N Engl J Med 2009;361:605–611.
Address correspondence to: John C. O’Horo, M.D. Aurora UW Medical Group Department of Graduate Medical Education 945 N 12th Street Milwaukee, WI 53233 E-mail: [email protected]
Evaluating Traditional Prognostic Measures in Patients Undergoing Hypothermia After Cardiac Arrest John C. O’Horo, M.D.,1 Mihail Andreev, M.D.,2 Wael Hassan, M.D.,1 and Asif Anwar, M.D., M.S., FCCP 3*
Background: Therapeutic hypothermia is one of the few interventions that improve mortality and neurologic outcomes in patients who have experienced cardiac arrest. There is a lack of validated tools to predict survival in patients who have undergone hypothermia after cardiac arrest (HACA). Methods: A retrospective analysis was performed of all patients who underwent HACA at Aurora St. Luke’s Medical Center (ASLMC) since the protocol was implemented in September 2008. Initial rhythm, whether percutaneous coronary intervention (PCI) was performed, lactate levels, duration of resuscitation, and APACHE II scores were compared for survivors and non-survivors, and a logit binary regression model was constructed. Results: A total of 143 patients were identified and had data abstracted. APACHE-II, duration of resuscitation, and initial rhythm were all strongly correlated with survival. Initial serum lactate levels were higher in nonsurvivors than survivors ( p = 0.005), though the trend of lactate change at next draw was not predictive. Quantitative TnI was not significantly different between arms. Conclusion: Lactate levels show promise as a biomarker for survival in HACA patients resuscitation length, presence of PCI, and APACHE-II scores can provide good prognostic information, even in the early hours following a resuscitation event.
demonstrated promise in the HACA population in three studies that drew lactate levels post-resuscitation. The first was an observational study, which noted a correlation between lactate levels and survival (Oddo et al., 2008). The second was a prospective study, which found that duration of resuscitation and lactate levels were the only factors significantly predictive of survival (Shinozaki et al., 2011). The final study, a multicenter retrospective study, measured lactate 1 hour after completing resuscitation, and found a stepwise association with lactate levels and mortality, with lactates > 10 mmol/L associated with 92% mortality (Cocchi et al., 2001). The HACA protocol currently used at Aurora St. Luke’s Medical Center (ASLMC), a 730-bed tertiary referral hospital in Milwaukee, has routinely determined lactate, troponin I (TnI), and APACHE-II data on all patients since approval of the local protocol in September 2008. This study retrospectively analyzes these data from this patient population.
Introduction
M
ore than 300,000 cardiac arrests occur each year in the United States, with a dismal overall survival rate of less than 10% (Nichol et al., 2008). Hypothermia after cardiac arrest (HACA) is one of the few interventions that has been convincingly demonstrated to improve mortality and neurologic outcomes in patients who have experienced cardiac arrest (Bernard et al., 2002; HACA Study Group, 2002). In spite of this, mortality remains high, and it remains difficult to predict which patients will benefit from the procedure. Critical care physicians are asked to provide a prognosis for HACA patients. Unfortunately, this is quite difficult due to a lack of validated tools. Traditional neurologic assessments do not always provide adequate prognostic information (Young, 2009; Bisschops et al., 2011), and studies into traditional biomarkers and scoring tools have largely not been tested in the HACA population. Acute physiology and chronic health evaluation II (APACHE-II) score and lactate levels have both been shown to be predictive in the general cardiac arrest population (Schuetz et al., 2010; Cocchi, 2011) and lactate levels specifically have
Methods ASLMC implemented a HACA protocol in September 2008 (see Fig. 1). This protocol underwent no changes during the
1
Aurora UW Medical Group, Milwaukee, Wisconsin. Aurora Medical Group, Department of Hospital Medicine, Milwaukee, Wisconsin. 3 Aurora Medical Group, Department of Critical Care, Milwaukee, Wisconsin. *Current additional affiliations: University of Wisconsin School of Medicine and Public Health, Madison, Wisconsin; United Health Systems, Kenosha, Wisconsin. 2
24
PROGNOSIS TOOLS IN HYPOTHERMIA AFTER CARDIAC ARREST
25
Statistical analysis was performed using Minitab 15 software. Correlation between survivorship was independently tested for each variable using one-way analysis of variance (ANOVA) testing. Several logit binary regression models were constructed from these data for comparison of the relative importance of each variable. Results
FIG. 1. Interval plot of survival by year. No statistically significant difference was seen ( p = 0.34). Mean survival followed the linear regression 175.47–0.0767(x) where x was the year of the code (R2 = 0.96).
One hundred and forty-four patients were started on the HACA protocol at ASLMC between September 2008 and April 2011 and included in the study. Of these, 91 (63.2%) were categorized as non-survivors, with 84 (58.3%) patients expiring and seven (5%) being transferred to a hospice. The study identified 53 (37.7%) survivors with 29 (20%) being discharged home, five (3.5%) to a long-term care facility, 15 (10%) to rehab, two (1.4%) to a skilled nursing facility, and one (0.7%) leaving against medical advice. Trend data
study period. Since the initiation of this protocol, patients undergoing therapeutic hypothermia have been tracked in a database. Following institutional review board approval, patient names and admissions were obtained from this database, and accessed in the electronic medical record system. Data were abstracted by three of the researchers independently. All information was abstracted twice, and conflicts in the resultant data were resolved with discussion among the researchers. Subjects were first categorized by their discharge disposition, with those who expired in-house or transferred to hospice being categorized as non-survivors, and the rest as survivors. On each patient, the rhythm at initiation of resuscitative efforts and duration of resuscitation was abstracted from a quality database maintained for HACA patients. Rhythms were coded as able to be defibrillated (ventricular tachycardia and fibrillation) and unable to defibrillate (asystole and pulseless electrical activity). Patients who were shocked with an automatic external defibrillator in the field without presenting rhythm were coded as able to defibrillate. Data on time down before resuscitation were initiated on out-of-hospital arrests were not available. Other data were abstracted from the electronic medical record system. Procedure notes were reviewed on all patients who underwent coronary catheterization. Those who underwent percutaneous coronary intervention (PCI) were noted as such in the data abstraction. First lactate level drawn by the lab was abstracted for each patient who had one drawn within 24 hours of code event. Point-of-care and bedside testing values were excluded because of a high variability seen in these, and questions about their quality at this institution. When possible, second lactate and the rate of change of lactate were also abstracted. First quantitative TnI were also abstracted. Originally, procalcitonin levels were set to be abstracted as well, but no HACA patients had this level checked. Finally, APACHE-II scores were calculated for each patient and determinants abstracted from the 24 hours of data preceding cooling. For patients in hospital less than 24 hours before the cooling was started, the most abnormal set of labs and vitals between admission and cooling was abstracted. The final APACHE-II score was calculated using the tool at www.sfar.org/scores2/apache22.html.
HACA patients were stratified by year to analyze if there was a significant difference in survival over the study period. In 2008, 11 patients were started on HACA and 45.5% survived. In 2009, 53 HACA protocols were initiated and 41.5% survived. The year 2010 saw 61 HACA patients with a 31.2% survival rate. Although a linear regression model demonstrated an 8% drop in survival each year (R2 = 0.96), this trend was not significant by ANOVA ( p = 0.34). This is illustrated in Figure 1. Initial rhythm and PCI Initial rhythm was significantly different between survivors and non-survivors. Asystole had the lowest survival and ventricular fibrillation and tachycardia were tied for best survival ( p < 0.001). When coded as shockable versus not, shockable rhythms had significantly better survival than nonshockable rhythms ( p < 0.001). This is depicted in Figure 2. Thirty-six patients underwent PCI after their cardiac arrest, with a survival rate of 55.6% compared with 29.2% in those who did not have percutaneous coronary intervention ( p = 0.004).
FIG. 2. Survival rates by rhythm. Better survival was seen in rhythms able to be defibrillated. VF and VT patients had a combined 52.3% survival rate compared to 10.7% in combined PEA/asystole patients ( p < 0.001).
26
O’HORO ET AL.
Lactate levels and rate of change Eighty-eight (62%) subjects had a first lactate level drawn. Time to first draw after admission was not different between survivors and non-survivors ( p = 0.07). First lactate level differed significantly ( p = 0.005) between arms, with a mean level of 3.4 mmol/L (SD = 2.1 mmol/L) in survivors and 5.9 mmol/ L (SD = 4.5 mmol/L) in non-survivors. Second lactate levels were drawn on 49 (34%) patients, also without significantly different times to draw ( p = 0.752). Again, results failed to achieve statistical significance, with a mean level of 3.4 mmol/L (SD = 1.9 mmol/L) in survivors, and 4.1 mmol/L (SD = 3.8 mmol/L) in non-survivors ( p = 0.53). Rates of lactate change were calculated for all 49 patients with a second lactate. Results again failed to achieve significance with a mean change of - 0.15 mmol/L per hour (SD = 0.37 mmol/L per hour) in survivors and - 0.27 mmol/L per hour (SD = 0.76 mmol/L per hour) in non-survivors ( p = 0.59). Quantitative TnI Quantitative TnI level was determined in 125 patients. Levels were not significantly different between groups ( p = 0.481); survivors averaged 17.37 ng/mL (SD = 55.44 ng/ mL) while non-survivors averaged 26.01 ng/mL (SD = 55.33 ng/mL). The validity of this observation is questionable in light of significantly different draw times, with survivors having TnI levels drawn 7.4 hours after code versus nonsurvivors 3.6 hours after code ( p = 0.03). Duration of resuscitation Duration of resuscitation was available on code documentation for 121 patients. Survivors had significantly shorter resuscitation times, averaging 18.3 (SD = 20.1) minutes versus 34.0 (SD = 27.6) minutes in non-survivors ( p = 0.001). APACHE II data Individual factors in the APACHE-II score are detailed in Table 1. A strong correlation was noted with mean arterial pressure, pH, creatinine, hematocrit, and Glasgow Coma Scale, though the aggregated APACHE-II score was the most predictive by far. Paired t-tests and Pearson correlation coef-
ficients demonstrate that the APACHE-II predicted death rate is similar to what is observed in this study (see Fig. 2). Binary regression modeling Several regression models were developed using combinations of statistically significant variables (see Table 2). These results were consistent with the ANOVA analysis. Length of resuscitation, lactate levels, and APACHE II score were negatively associated with survival, while rhythms subject to defibrillation and undergoing PCI were associated with decreased mortality. The relative contributions to each of these under the PCI/ duration of resuscitation/first lactate/APACHE II/Initial Rhythm model is modeled in Figure 2. This model demonstrated a good correlation with observed variables, with a Sommer’s D statistic of 0.85 and a Hosmer–Lemeshow statistic of 0.775, failing to reject the hypothesis that the model does not fit the data. All attempts at modeling component variables of the APACHE II score demonstrated inferior predictive value relative to the unaltered APACHE II score (see Fig. 3). Likewise, models including the TnI, the rate of lactate change, and second lactate all decreased the correlation with observed data. Marginal effects were tested with the model, demonstrating linear relationships with the variables in the model. The most significant relative variable was duration of resuscitation, followed by APACHE score with a relatively minor contribution of lactate levels. The categorical variables, PCI, and ability to defibrillate had the most impact on the predicted outcome. Discussion Although subject to limitations inherent to its retrospective design, this study identifies several useful predictors for survival in the minutes following the decision to initiate hypothermia protocol. In addition, this also illustrates the need for caution when interpreting the prognostic value of lactate for the HACA population. Although lactate levels did achieve statistical significance in this and other studies (Young, 2009; Bisschops et al., 2011), the logistic regression models demonstrate that lactate is a relatively minor contributor compared to APACHE II levels,
Table 1. Characteristics of Survivors Versus Non-survivors Variable Age (years) APACHE-II score Lactate (mmol/L) Proportion in-house arrest Proportion male Duration of resuscitation (min)
Mean 63.4 60.4 29.8 25.8 6.92 5.51 0.10 0.13 0.63 0.78 34.0 18.4
(non-survivors) (survivors) (non-survivors) (survivors) (non-survivors) (survivors) (non-survivors) (survivors) (non-survivors) (survivors) (non-survivors) (survivors)
SD
p value
14.8 16.7 6.5 6.7 4.32 3.26 0.31 0.34 0.05
0.264
27.6 20.6
0.002
0.001 0.092 0.625 0.08
P values are calculated using a one-way ANOVA model. APACHE-II score was calculated using the Socie´te´ Franc¸aise d’Anesthe´sie de Re´animation APACHE II calculator at www.sfar.org/scores2/apache22.html. SD, standard deviation.
PROGNOSIS TOOLS IN HYPOTHERMIA AFTER CARDIAC ARREST
27
Table 2. Logistic Regression Models Model Lactate predictive model
Lactate/TnI/APACHE II model
Duration of resuscitation/ rhythm/PCI model
APACHE constituent model
Restricted APACHE model
Restricted APACHE score contribution model
Duration of resuscitation/ Able to defibrillate/PCI model PCI/ Duration of resuscitation/ Lactate/ APACHE II/ Initial Rhythm
PCI/
Logistic regression model variables included Constant First lactate Next lactate Rate of lactate change First lactate2 Constant First lactate TnI APACHE-II score Constant Initial rhythm PEA VT VF PCI Duration of resuscitation Constant Temperature MAP HR RR FiO2 PaO2 pH Na K Creatinine ARF present HCT WBC GCS Age Chronic disease Constant MAP pH Creatinine Hct Constant MAP APACHE contribution pH APACHE contribution Hematocrit APACHE contribution Constant PCI Shockable rhythm Duration of resuscitation Constant Duration of resuscitation PCI Lactate APACHE-II Initial Rhythm PEA VF VT Constant
Variable p values
Hosmer–Lemeshow goodness of fit test
Sommer’s D statistic
- 7.26 - 0.025 - 0.067 0.261 0.0003 1.047 - 0.224 - 0.005 - 0.020 - 2.058 1.969 2.969 3.781 1.613 - 0.085
0.136 0.920 0.759 0.826 0.158 0.374 0.031 0.431 0.634 0.073 0.134 0.001 0.023 0.016 0.001
0.444
0.28
0.189
0.39
0.691
0.79
- 34.1 0.043 0.009 0.010 - 0.021 - 0.004 - 0.002 2.92 0.047 0.079 - 0.665 - 0.21 0.031 0.013 0.143 - 0.001 - 0.031 - 18.9 0.010 2.34 - 0.660 0.438 0.080 0.064 - 0.232 - 0.373 - 0.948 1.372 2.564 - 0.0868 - 19.14 - 0.068 1.177 - 0.024 - 0.032
0.019 0.547 0.286 0.268 0.579 0.789 0.293 0.044 0.242 0.736 0.061 0.608 0.288 0.699 0.237 0.967 0.960 0.037 0.222 0.062 0.033 0.107 0.808 0.651 0.046 0.109 0.146 0.019 0.000 0.000 0.997 0.018 0.280 0.854 0.592
0.483
0.51
0.592
0.42
0.669
0.31
0.500
0.78
0.775
0.85
19.98 22.09 20.80 - 0.677
0.997 0.997 0.997 0.699
0.069
0.81
Coefficient
(continued)
28
O’HORO ET AL. Table 2. (Continued) Logistic regression model variables included
Model Duration of resuscitation/ Lactate/ APACHE II/ Able to defibrillate
PCI Duration of resuscitation Lactate APACHE II Able to Defibrillate
Coefficient
Variable p values
1.060
0.274
- 0.063 - 0.142 - 0.003 2.988
0.020 0.237 0.000
Hosmer–Lemeshow goodness of fit test
Sommer’s D statistic
Coefficients for logit function, p values, and relative goodness of fit for all models tested shown. PCI, percutaneous coronary intervention.
FIG. 3. Dotplot of APACHE II score for survivors versus non-survivors. Paired t-test comparing to APACHE-II predicted death rate shows a high degree of similarity ( p = 0.631). Pearson correlation coefficient = 0.777 ( p = 0.0000).
duration of resuscitation, and ability to defibrillate rhythm. Lactate levels, though useful, should only be interpreted within the context of these other factors. Duration of resuscitation was highly predictive of survival, with longer times to spontaneous return to circulation being strongly correlated with higher mortality. This is consistent with other studies into resuscitation (Rosenberg et al., 1993). The utility of the APACHE II score was remarkable. Despite several attempts at modeling, no combination of the component variables was more predictive than the APACHE score itself. Previous data have demonstrated their predictive value in post-resuscitation patients (Bialecki and Woodward, 1995), but it has never been validated in the HACA population specifically before. A previous study has found the degree of quantitative TnI elevation to correlate with survival ( Joly et al., 2011). This was not replicated in this study. Several factors may have contributed to this. The TnI was drawn significantly earlier, on average, in patients who did not survive. Given the rapid peaking and declining of TnI, this severely limits the validity of comparing these data. Another factor is the hospital itself; ASLMC is a cardiology referral center with a 24-hour cardiac catheterization lab. Many of the STEMI related cardiac arrest patients were brought to ASLMC to receive coronary intervention. These patients would tend to have higher TnI, and the predictive effect of this would be blunted by the benefit of PCI. Several other biomarkers have been proposed as tools in predicting survival in HACA patients, including procalcitonin, cell-free plasma DNA levels (Amalich et al., 2010), neuron specific enolase, and S-100b (Sanfilippo et al., 2010). None of the patients in the study period had any of these values determined. Future studies to assess these potential prognostic tools are needed. The overall survival rate seen in HACA patients at ASLMC was 37.7%. This is significantly lower than reported mean of 65% seen in a review article (Sagalyn et al., 2009). The exact reason for this is unclear. Future multicenter trials would
better be able to determine the relative predictive value of these factors in HACA patients. Despite the limits inherent to a retrospective study, these data can be of value to clinicians. When providing early prognostic information for HACA patients, APACHE-II criteria and length of resuscitation efforts can provide useful tools. Lactate levels, while useful, should be interpreted with caution, pending further prospective study into this additional other biomarkers. Acknowledgments The authors would like to thank Sara Marzinski, RN, for providing access to the hypothermia protocol patient database. Disclosure Statement No competing financial interests exist. References Amalich F, Mene´ndez M, Lagos V, Ciria E, Quesada A, Codoceo R, Vazquez JJ, Lo´pez-Collazo E, Montiel C, et al. Prognostic value of cell-free plasma DNA in patients with cardiac arrest outside the hospital: an observational cohort study. Crit Care 2010:14:R47. Bernard SA, Gray TW, Buist MD, Jones BM, Silvester W, Gutteridge G, et al. Treatment of comatose survivors of out-ofhospital cardiac arrest with induced hypothermia. N Engl J Med 2002;346:557–563. Bialecki L, Woodward RS. Predicting death after CPR. Experience at a nonteaching community hospital with a full time critical care staff. Chest 1995;108:1009–1117. Bisschops LL, van Alfen N, Bons S, van der Hoeven JG, Hoedemaekers CW. Predictors of poor neurologic outcome in patients after cardiac arrest treated with hypothermia: a retrospective study. Resuscitation. 2011;82:696–701. Cocchi MN, Miller J, Hunziker S, et al. The association of lactate and vasopressor need for mortality prediction in survivors of cardiac arrest. Minerva Anestesiol 2001;77:1063–1071.
PROGNOSIS TOOLS IN HYPOTHERMIA AFTER CARDIAC ARREST Cocchi MN, Miller J, Hunziker S, Carney E, Salciccioli J, Farris S, et al. The association of lactate and vasopressor need for mortality prediction in survivors of cardiac arrest. Minerva Anestesiol 2011;11:1063–1071. Hypothermia After Cardiac Arrest Study Group. Mild therapeutic hypothermia to improve the neurologic outcome after cardiac arrest. New Engl J Med 2002;346:550–556. Joly SS, Shenkman H, Brieger D, Fox KA, Yan AT, Eagle KA, Steg PG, Lim KD, Quill A, Goodman SG, GRACE investigators. Quantitative troponin and death, cardiogenic shock cardiac arrest and new heart failure in patients with non-STsegment elevation acute coronary syndromes (NSTE ACS): insights from the Global Registry of Acute Coronary Events. Heart 2011;97:197–202. Nichol G, Thomas E, Callaway CW, Hedges J, Powell JL, Aufderheide TP, et al. Regional variation in out-of-hospital cardiac arrest incidence and outcome. JAMA 2008;300:1423–1431. Oddo M, Ribordy V, Feihl F, Rossetti AO, Schaller MD, Chiolero R, et al. Early predictors of outcome in comatose survivors of ventricular fibrillation and non-ventricular fibrillation cardiac arrest treated with hypothermia: A prospective study. Crit Care Med 2008;36:2296–2301. Rosenberg M, Wang C, Hoffman-Wilde S, Hickam, D, et al. Results of cardiopulmonary resuscitation. Failure to predict survival in two community hospitals. Arch Intern Med 1993;154:1370–1375.
29
Sagalyn E, Band RA, Gaieski DF, et al. Therapeutic hypothermia after cardiac arrest in clinical practice: review and compliation of recent experiences. Crit Care Med 2009;37:223–226. Sanfilippo F, Li Volti G, Ristagno G, Murabito P, Pellis T, Astuto M, Gullo A, et al. Clinical biomarkers in brain injury: a lesson from cardiac arrest. Front Biosci (Schol Ed) 2010;2623–2640. Schuetz P, Affolter B, Hunziker S, Winterhalder C, Fischer M, Balestra GM, et al. Serum procalcitonin, C-reactive protein and white blood cell levels following hypothermia after cardiac arrest: a retrospective cohort study. Eur J Clin Invest 2010;40:376–381. Shinozaki K, Oda S, Sadahiro T, Nakamura M, Hirayama Y, Watanabe E, et al. Blood ammonia and lactate levels on hospital arrival as a predictive biomarker in patients with out-ofhospital cardiac arrest. Resuscitation 2011;82:404–409. Young GB. Clinical practice. Neurologic prognosis after cardiac arrest. N Engl J Med 2009;361:605–611.
Address correspondence to: John C. O’Horo, M.D. Aurora UW Medical Group Department of Graduate Medical Education 945 N 12th Street Milwaukee, WI 53233 E-mail: [email protected]
