Influence of Comorbidity on the Outcome of Patients Treated for Out-of-Hospital Ventricular Fibrillation
Background A number of factors have previously been shown to be predictive of survival from out-of-hospital ventricular fibrillation. These include witnessed collapse, prompt initiation of cardiopulmonary resuscitation, early application of defibrillation, and younger age. Arrests occurring away from home are also associated with improved survival. Additionally, hospital mortality after successful resuscitation has been related to a history of congestive heart failure as well as to some of the factors noted above. An association of prearrest comorbidity with outcome has not been systematically evaluated.
Methods and Results We define here a comorbidity index, which is constructed from histories of chronic conditions as well as a number of recent symptoms in 282 victims of out-of-hospital VF. This indicator of comorbidity is strongly associated with outcome (P=.004). However, when analyzing a comprehensive set of predictors of survival after out-of-hospital ventricular fibrillation, including the index of comorbidity, we could identify overall only about one fourth of the variation that one might hope to account for.
Conclusions Comorbidity appears to be an important (but usually overlooked) predictor of survival from out-of-hospital ventricular fibrillation. However, most of the statistical variability in predicting survival remains unexplained when we consider comorbidity in conjunction with previously identified predictors of survival.
A number of factors predictive of survival from out-of-hospital ventricular fibrillation have been reported.1 2 3 4 5 These include prompt initiation of cardiopulmonary resuscitation (CPR) (typically when the collapse has been witnessed and when an efficient method for requesting emergency medical services is used), prompt application of defibrillatory shock, younger age, and location of episode. Recently, race and socioeconomic status have also been shown to be associated with survival.6 7 8 For patients who are resuscitated, failure to survive the subsequent hospitalization has been associated with a history of congestive heart failure as well as a number of the variables mentioned above.9 Since an evaluation of the treatment of cardiac arrest (or any disorder) must consider competing factors for recovery, an appreciation of all factors affecting outcome is desirable. Thus, in an effort to further clarify the determinants of outcome, we evaluated a comorbidity index derived from histories of chronic conditions as well as symptoms before arrest. Intuition would probably lead one to predict that patients with preexistent comorbidity would fare less well than victims who have been more healthy. However, to the best of our knowledge, no analyses of that relationship have been reported in patients treated for out-of-hospital cardiac arrest.
Seattle Emergency Medical Services System
The Seattle Fire Department’s emergency care system has operated for more than 25 years, providing a two-tiered response for all of the city’s out-of-hospital medical emergencies. Basic and advanced life support units are dispatched simultaneously whenever a life-threatening emergency such as cardiac arrest is perceived to be present. Seattle’s population in 1990 was 516 000 persons, of whom 417 000 (81%) were ≥20 years old. During the years 1989 and 1990, the annual rate at which cardiac arrest (attributed to heart disease) was treated by the Seattle paramedics was 1.04 per thousand persons ≥20 years old. The characteristics and outcomes of patients treated by the system have been described.2 3 7
Between May 1986 and August 1988, we conducted a study to evaluate alternative dialogues for providing instruction over the telephone for CPR. During this time, extensive data were collected for victims of out-of-hospital cardiac arrest in Seattle in whom ventricular fibrillation was the first recorded rhythm. We excluded episodes obviously not due to underlying heart disease, eg, electrocution, drug overdose, and near-drowning. Because the interviews we performed were a component of a study relating to CPR instruction by telephone, we also excluded cases in which the caller would not be able to provide CPR (eg, relayed calls or instances in which the victim was inaccessible). Additionally, cases were excluded if the cardiac arrest occurred after the call to 911. A total of 356 cases were considered eligible for this study.
Telephone interviews of the caller (or other witness) were carried out, on average, 7 weeks after the arrest. Information was sought concerning histories relating to chronic problems, including use of heart medications, previous heart attack, high blood pressure, chest pain (or angina), heart failure, chronic pulmonary disease, diabetes, cancer, gastrointestinal disorders, and other chronic conditions. We also elicited symptoms that had occurred within 2 days of the collapse: chest pain, dizziness or faintness, indigestion, shortness of breath, nausea, fatigue, or weakness. We determined whether the patient had visited a doctor or medical facility within 2 days, and specific inquiry was made regarding prior heart surgery.
To develop a measure of comorbidity, we first calculated two simple proportions that were our a priori choices for summary descriptions. We computed (1) a chronic factor (CF)=(number of positive+1/2 number of unknown)/total number of chronic conditions and (2) a symptom factor (SF)=(number of positive symptoms+1/2 number of unknown)/total symptoms. We used logistic regression analysis10 to investigate the relation of each of these factors to outcome, to examine whether a history of heart surgery or the occurrence of a physician visit before the episode provided additional information, and finally to construct a single comorbidity index (see below). The deviance explained by the model is the ratio of the deviance based on the model with predictors compared with a model with no predictors.11
Symptoms and Chronic Histories
Characteristics of the patients are shown in Tables 1⇓ and 2⇓. The mean age was nearly 66 years; on average, approximately 4 of the 10 chronic histories and 1.5 of the 6 recent symptoms were reported as present. The use of heart medications was acknowledged for 60% of the patients. Overall, 32% survived through hospital discharge.
We were able to acquire comorbidity data for 282 (79%) of the 356 episodes (Table 2⇑). As expected, this information was frequently ascertainable when the caller was related to the patient (225 of 259, 87%). It was not possible to obtain an interview for 22 cases; in 52 others, the respondent knew nothing of the patient’s history and recent symptoms.
There was an association between the number of recent symptoms and the likelihood of a physician visit before the episode (Fig 1⇓). Both a history of a recent physician visit and the number of recent symptoms were associated with decreased survival (Fig 2⇓); however, the occurrence of a physician visit provided no additional ability to predict survival over that of the summary proportion of symptoms. Similarly, the contributions of the symptom factor were not increased when the timing based on the most recent symptom was taken into consideration, ie, less than 10 minutes, 10 to 60 minutes, or 1 hour to 2 days.
The chronic factor was also predictive of survival (P=.006), but there was no interaction or additional effect of a history of heart surgery when survival was analyzed with the chronic factor as a covariate.
The rank correlation between the CFs and SFs was 0.22 (P<.001). Since the two variables did not interact significantly for predicting outcome, a comorbidity index determined by logistic regression was a simple linear combination: 1.67×CF+SF. The comorbidity index averaged 1.01 (SD, 0.43). It was significantly lower in patients who survived compared with those who died (0.87 versus 1.08, P<.0005) and was weakly (r=.21) but significantly (P<.001) correlated with age. As shown in Table 3⇓, the comorbidity index was strongly related to the location of collapse (comorbidity lowest in episodes away from home; P<.0005) and to activity (lowest in episodes associated with high activity levels; P<.0006). Emergency medical service response times were not related to the comorbidity index.
In a logistic regression analysis, the comorbidity index contributed significant additional risk information above that provided by the previously identified factors known to affect outcome (Table 4⇓; P<.004). In a logistic model with no forced variables, the comorbidity index was the first variable selected (<.0001).
We have shown that comorbidity is an important predictor of survival in patients treated for out-of-hospital ventricular fibrillation—in this analysis, the most powerful predictor. The comorbidity index described here is a simple combination of the proportion of positive or unknown histories regarding 10 chronic conditions (heart failure, myocardial infarction, use of heart medications, diabetes, hypertension, chest pain, chronic pulmonary disease, gastrointestinal disorders, cancer, and other chronic conditions) and the proportion of positive or unknown recent (within 2 days) symptoms: chest pain, dizziness, indigestion, dyspnea, nausea, and fatigue.
The comorbidity index is largely independent of the other predictors that have been reported. However, even with the inclusion of the comorbidity index, the totality of known predictors contributes only a little more than 10% of the discrimination that would be provided by perfect prediction (Fig 3⇓) and only about 25% of what might be expected with a realistically ideal model (based on simulations not reported here). Nevertheless, even this modest accounting can be useful for group predictions, as is demonstrated in Table 5⇓, in which actual survival rates are tabulated for quartiles of predicted risk.
Although in our model the comorbidity index was selected as the variable most predictive of outcome, it could be argued that the more obvious and previously noted predictors should be included in the model first. In such an analysis, the comorbidity index still enters the model very significantly (P<.004).
In conclusion, we have demonstrated that comorbidity was an important predictor of survival from out-of-hospital ventricular fibrillation in a reasonably large set of patients. We have also noted that the state of the art for predicting outcome is far from perfect, probably related to factors that are unrecognized but also to limitations of the present data, eg, temporal uncertainties before the 911 call.
Although the observations reported here are quite probably relevant to comparable emergency medical care systems, that supposition requires validation. We would also suggest that major changes in the delivery of care could alter the significance of the findings. For example, in a system in which many responses were much more rapid, ie, 1 to 2 minutes, it is possible that response time would be a more powerful predictor and that a greater proportion of the deviance might be explained.
This study was supported in part by grants from the Agency for Health Care Policy and Research, the Medic One Foundation, and the Alfred and Tillie Shemanski Trust Fund.
Reprint requests to Leonard A. Cobb, MD, Division of Cardiology, Box 359748, Harborview Medical Center, 325 Ninth Ave, Seattle, WA 98104. E-mail email@example.com.
- Received July 19, 1995.
- Revision received November 14, 1995.
- Accepted November 19, 1995.
- Copyright © 1996 by American Heart Association
Thompson RG, Hallstrom A, Cobb LA. Bystander-initiated cardiopulmonary resuscitation in the management of ventricular fibrillation. Ann Intern Med. 1979;90:737-740.
Weaver WD, Cobb LA, Dennis D, Roy R, Hallstrom AP, Copass MK. Amplitude of ventricular fibrillation waveform and outcome after cardiac arrest. Ann Intern Med. 1985;13:927-929.
Larsen MP, Eisenberg MS, Cummins RO, Hallstrom AP. Predicting survival from out-of-hospital cardiac arrest: a graphical model. Ann Emerg Med. 1993;22:9-15.
Cox DR. Analysis of Binary Data. London, UK: Chapman & Hall Ltd; 1977.
McCullagh P, Nelder JA. Generalized Linear Models and Monographs on Statistics and Applied Probability. London, UK: Chapman & Hall Ltd; 1983.