Pre-arrest predictors of survival after resuscitation from out-of-hospital cardiac arrest in the elderly a systematic review

We searched MEDLINE with an extensive search strategy to identify studies published between January 1980 and May 2011 that investigated prognostic factors for survival of out-of-hospital CPR (Additional file 1). In addition, we checked the reference lists of the selected studies to identify missing relevant articles and we used a multidisciplinary Dutch guideline about decision-making on resuscitation in older patients as an additional source of studies [9]. For this guideline, MEDLINE, Embase, Web of Science, CINAHL, Cochrane DSR, DARE and Cochrane Controlled Trial Register and DARE were searched to identify studies published between 1950 and 2008.

The root search was a combination of synonyms for cardiopulmonary resuscitation ([cardiopulmonary resuscitation] OR [CPR] OR [mouth to mouth]) and search terms for cardiopulmonary arrest ([heart arrest] OR [cardiac arrest] OR [cardiopulmonary arrest] OR [sudden death]) combined with outcomes ([quality of life] OR “cerebral recovery” [tiab] OR [functional impairment] OR “hospital discharge” [tiab]). All search terms were entered as free text words and as Medical SubHeadings (MeSH-terms). To limit the results to the geriatric population, we used a sensitive filter for geriatric medicine [10]. The search strategy is available through the authors.

First, one author (EvdG) selected studies based on the titles and abstracts. Then, two researchers (EvdG and FvdW) screened the full texts of the remaining articles more thoroughly. Disagreements were discussed with a third reviewer (LH). Only studies that were written in English were included.

For this review, we included studies that investigated patients who required CPR for a cardiopulmonary arrest in an out-of-hospital setting (including nursing homes). We defined cardiopulmonary arrest as the sudden cessation of spontaneous circulation and respiration leading to loss of consciousness and death when CPR is not provided. CPR was defined as the use of chest compressions and rescue breathing, with or without advanced life support, delivered according to the protocols that were applied in the study period.

We included studies in which the mean age of the participants was 70 years or more or in which different age groups were presented separately, including patients aged 70 years or more. Eligible studies assessed ‘age’ as a clinical predictor of survival or mortality after CPR, or as predictor for the quality of life of survivors, both univariably and multivariably. Studies also had to report ‘survival to discharge’ as a main outcome measure; studies that only reported ‘recovery of spontaneous circulation (ROSC)’ or ‘hospital admission’ were excluded. We excluded studies that described patients with loss of consciousness due to seizure, sole respiratory arrest or cardiopulmonary arrest due to trauma or drowning.

To assess the methodological quality of the included studies, we used a checklist based on the checklist developed by Hayden et al. [11]. This checklist assesses six domains of bias in a systematic review of prognosis studies (Table 1). Each item could score ‘low risk of bias’, ‘moderate risk of bias’, ‘high risk of bias’ or ‘unknown risk of bias’.

Two researchers (EvdG and FvdW) independently performed the quality assessment. When necessary, disagreements were resolved through discussion with a third reviewer (LH).

We used a standardized form to collect information on patients’ demographic and arrest characteristics. Furthermore, we extracted data on survival to discharge and beyond and on the quality of life, cognitive and functional status of survivors if reported. The reported arrest characteristics are all known to influence the outcome of cardiopulmonary resuscitation and can be considered as confounders for which analyses should be adjusted [3, 12].

The outcomes ‘survival to discharge’ and’30-day survival’ were combined into short-term survival. Other outcome measures, such as long-term survival, quality of life and functional dependence of survivors, were reported separately.

For studies that exclusively included participants aged 70 years or above or reported data on a subgroup with this age and were sufficiently similar with respect to the participant and arrest factors, we calculated a pooled overall survival rate. We used an exact likelihood approach based on the binomial within-study distribution. This model allows for zero survivors in one or more studies [13]. Because we expected substantial heterogeneity in the reporting of quality of life and functional dependence, we did not pool these results. Nor did we perform a meta-analysis of the prognostic accuracy of the various individual prognostic factors, since we expected substantial heterogeneity between the primary studies, for example in the number and type of covariates that were studied and, more importantly, in the predictors that were included in a multivariable adjusted analyses (if done) [14, 15].

The search resulted in 5,436 articles (Figure 1). Of the 132 potentially relevant articles, 22 were included. In addition, one additional record was identified through other sources. The main reasons for exclusion were a mean age below 70 or no separate subgroup with participants >70 years and the examination of in-hospital CPR only. From the large Swedish cohort study by Herlitz, from which many reports were published, we included one key publication that met the inclusion criteria [16].

The majority of the studies scored a low-to-moderate risk of bias on the items of study participation, study attrition and prognostic factor measurement (Figure 2).

When a high risk of bias was assessed for the domain of study participation, it was because important baseline characteristics such as comorbidity or age were missing. In addition, some studies addressed a specific patient group that did not match the current research question. For example, some studies only described witnessed arrests or only included patients who were admitted alive [17, 18]. The reasons for not reporting on the entire cohort of resuscitated patients in the analysis were not always listed. Therefore, it was not clear whether there were differences between the participants who were analyzed and those who were not, which could have introduced bias on the item of study attrition.

On the item of prognostic factor measurement, most of the studies were assessed as having a moderate risk of bias. Typically, the prognostic factor ‘age’ was described; however, other factors, such as co-morbidity or functional dependence, were often not reported. In these cases, the score on this domain was ‘moderate risk of bias’ at best.

In all cases the outcomes quality of life and functional and cognitive status of survivors were measured adequately and reliable, and therefore we assessed no risk on misclassification bias on this item.

In only a few cases, the items of confounding and analysis could be assessed as low risk of bias, because adjustment for response time, percentage bystander CPR and type of rhythm was generally poorly controlled. Furthermore, when a multivariate analysis was performed, only the variables that were statistically significantly associated with the outcome in the univariate analysis were typically presented, explaining the low score on the ‘analysis’ item.

Additional file 2: Table S2 shows the characteristics of the 23 included studies. The total number of included patients was 44,582, with an age range of 33–99 years. Of the studies, four exclusively included elderly patients [17, 19‐21]. In five of the studies, the mean age of the included patients was 70 years or above [22‐26]. Fourteen studies provided a subgroup of elderly patients. For these studies, only the proportion survival in the oldest group is presented in Additional file 2: Table S2 [16, 18, 27‐38].

Thirteen studies were performed in the USA, and eight in Europe. The study populations and registered pre-arrest characteristics varied across studies. All studies were retrospective cohort studies or chart reviews, with the exception of Ghusn [17], which was a case control study. All but one study [38] reported at least survival to discharge; seven reported long term outcomes as well [23, 25, 27, 28, 37, 38].

Fourteen studies were sufficiently clinically homogeneous to perform a meta-analysis for survival (Figure 3). For patients aged 70 years or older, the pooled overall survival to discharge was 4.1% (95% confidence interval (CI) 3.0-5.6%; range 0–9.0%) (Table 2). This was lower than the general survival, as reported by Sasson et al. (pooled survival 7.6% (95% CI 6.7-8.4%) [3].

There was substantial clinical and statistical heterogeneity and reporting of statistical methods was often inadequate. Performing a meta-analysis of odd ratio’s (ORs) was impossible due to a wide variety in the statistical methods used such as the adjustment factors and the statistical models. Table 2 shows that the chance of survival significantly decreased as age increased, both in univariate [16, 22, 27, 30, 32‐34, 38] and multivariate analyses [16, 22, 27, 30, 32‐34]. In the multivariate analyses, most studies included only arrest factors, such as ‘witnessed arrest’, ‘bystander arrest’ and ‘shockable rhythm’ in the model; thus, these studies do not clarify the influence of pre-arrest comorbidity and functional status. Only the study of Fabbri [27] analyzed the effect of pre-arrest co-morbidities on the chance of survival (see Table 2). However, this study did not adjust for arrest factors and only examined witnessed arrests. Of the 23 included studies, six studies investigated the predictive value of nursing home residency for decreased survival to discharge [17, 19, 20, 30, 32, 38] (Table 3). For this group, the absolute survival chances were low and ranged from 0–5.1%. One study showed that nursing home residency was significantly associated with a lower chance of survival to discharge (OR 0.14) [19], whereas Deasy et al. presented a significant OR of 0.26 that was adjusted for arrest factors [30]. Although there were limited studies these data show that the chance of survival for this group is lower.

Of the included studies, eleven reported on characteristics of survivors such as functional and cognitive status [18, 21, 23, 27‐29, 32, 33, 37, 38] (Additional file 2: Table S2). Quality of life of survivors was reported in only one study [25].

Two studies reported that 7.5% of the patients for whom resuscitation was attempted survived neurologically intact to one year; however, one of these studies only examined witnessed arrests [27], and the other did not specify this outcome for older patients [23]. In patients that did not regain and sustain vital signs in the field, only 0.6% survived to discharge neurologically intact [24]. Other studies that included only patients over 70 years showed that although the overall survival was low, the majority of the survivors displayed moderate to good cerebral performance [18, 25, 28, 32, 37]. The study of Pleskot et al. showed no difference between younger and older survivors in cerebral performance, but the number of survivors was insufficient to identify significant differences [28].

In the Horsted study, survivors rated two of the eight quality of life aspects of the SF-36 scale as significantly worse than the age-matched normative scores. However, no specification for age was made [25].