Longer than 2 hours to antibiotics is associated with doubling of mortality in a multinational community-acquired bacterial meningitis cohort

To optimally define the association between time to effective antibiotic therapy and clinical outcomes in adult community-acquired bacterial meningitis. A systematic review of the literature describing the association between time to antibiotics and death or neurological impairment due to adult community-acquired bacterial meningitis was performed. A retrospective cohort, multivariable and propensity-score based analyses were performed using individual patient clinical data from Australian, Danish and United Kingdom studies. Heterogeneity of published observational study designs precluded meta-analysis of aggregate data (I2 = 90.1%, 95% CI 71.9–98.3%). Individual patient data on 659 subjects were made available for analysis. Multivariable analysis was performed on 180–362 propensity-score matched data. The risk of death (adjusted odds ratio, aOR) associated with treatment after two hours was 2.29 (95% CI 1.28–4.09) and increased substantially thereafter. Similarly, time to antibiotics of greater than three hours was associated with an increase in the occurrence of neurological impairment (aOR 1.79, 95% CI 1.03–3.14). Among patients with community-acquired bacterial meningitis, odds of mortality increase markedly when antibiotics are given later than two hours after presentation to the hospital.

Individual patient data collection. In an attempt to combine all individual patient data from the identified publications, corresponding and senior authors were contacted by email on at least three occasions to seek their involvement in this study. Statistical analysis. Descriptive data were reported in standard formats. Binary outcome variables were analysed using univariate and multivariable logistic regression. Stepwise selection (backward and forward) was used to select important predictors of health outcomes from the full model.
To account for unobserved inter-study variability among patients, clinical outcomes within studies were treated as correlated random effects. Generalizsed Linear Mixed Models were constructed for binary outcome variables (multivariable mixed logistic regression) and were used to estimate the association between time to treatment (antibiotic therapy) in hours and mortality (GOS = 1) and neurological impairment (GOS = 2-4), separately adjusting for other covariates. The associations were presented as adjusted odds ratio (aOR) for outcomes associated with time to antibiotics and selected covariates.
Propensity scoring was used to select a cohort of CABM patients matched for age and gender that had the same logistic probability of receiving treatment within a specific time. Exact matching for gender and age was undertaken utilising one-to-one matching using nearest neighbour matching within a calliper of 0.25.
Statistical analysis was conducted in SAS 9.4 (SAS Inc. Cary NC) and R software version 3.4.2 11 .

Results
Database searches found 778 publications matching search terms. Following removal of duplicates, title and abstract screening and full-text review, 18 publications were suitable for qualitative analysis (see Table 1 and Figure S1 PRISMA 9 diagram). Of these, eight studies reported effect estimates of CABM outcomes 6,[12][13][14][15][16][17][18] in relation to time to antibiotics and were considered for possible meta-analysis. Seven of these eight studies showed an association between increasing TTA and poor outcomes. Heterogeneity of published observational study designs precluded meta-analysis of aggregate data (I 2 = 90.1%, 95% CI 71.9-98.3%). This was due to marked differences in reporting of time function between admission and www.nature.com/scientificreports/ antibiotic administration and reported measures of clinical outcome association as described in Table 1. Additionally, interval to the measurement of mortality and neurological impairment, methods of data analysis, and predictive co-variables differed across the studies. A further ten studies 8,19-27 that described data on TTA but did not calculate associated outcomes were suitable for individual patient analysis.
The Townsville University Hospital CABM cases. Eighty-three patients from the period 1/1/2006 to 30/6/2018 with the specified ICD-10-AM codes for bacterial meningitis were identified. Twenty-five of these patients met the inclusion criteria. Patient characteristics, aetiological agents, time to effective antibiotics, and clinical outcomes are listed in Table 2 together with data made available by other authors for the individual patient outcome analysis. Individual patient data were shared from previous studies by Koster-Rasmussen et al. 6 and Stockdale et al. 8 . Dr Bodilsen provided updated data 7 from a recent study, Bodilsen et al. 18 , which was not part of our systematic review. Individual patient data were also made available from another study 14 where patients were treated immediately on arrival at a tertiary hospital after a variable time to CABM diagnosis at referring hospitals. The time from referring hospital presentation to antibiotic administration were not recorded, so these individual patient data could not be used. Institutional review board approval had been granted for the collection of all these data [6][7][8]18,26 . Other authors attempted to share individual patient data but were unable to retrieve these due www.nature.com/scientificreports/ to them being stored on redundant media 17,26 . The remaining authors did not reply to repeated invitations or declined to share data. Data describing 659 individual patients were available for analysis from the three previously reported hospitalbased cohorts 6-8 and the newly described TUH CABM cases. Unpublished data, were provided by Dr Miner (personal communication) as that included only patient age, time to antibiotics and outcomes. These data were collected with IRB approval, as a pilot study for a later publication 26 .
Demographic, clinical, microbiological characteristics, TTA as well as outcomes among studies providing individual patient data available are summarised in Table 2 and Fig. 1. In-hospital death and neurological impairment were described in TUH data and by Bodilsen et al. 7 Death and neurological impairment due to CABM were measured at 30 days. Stockland et al. 8 only reported death and not neurological impairment due to CABM. Outcomes due to CABM in the combined dataset were: death rate (21.6%, 95% CI 18.9-24.2) and frequency of neurological impairment (32.4%, 95% CI 30.0-36.0). A total of 563/659 (85.4%) patients received antibiotic therapy within the first 12 h of admission with a mortality of 103/563 (18.3%). Among patients with TTA > 12 h mortality was 39/96 (40.6%).
The most common organisms among the CABM patients studied were Streptococcus pneumoniae and Neisseria meningitidis. In both the Danish studies 6,7 a considerable proportion of CABM (14.6% 7 and 16.8% 6 ) were due to streptococcal species, Enterococcus faecalis, and Listeria monocytogenes. A further group of infrequent bacterial causes of CABM (Escherichia coli, Klebsiella pneumonia, Pasteurella multocida, and Capnocytophaga spp.) are referred to as 'other Gram-negatives' .
Analysis of the combined CABM data showed that the probability of mortality or neurological impairment was associated with increasing TTA (Table 3 and Fig. 2). In this cohort of patients, the probability of death following hospital admission was 11.5% (95% CI 8.0-15.1%) and the probability of neurological impairment was 27.8% (95% CI 21.7-33.8%) (Fig. 2). After a delay of 12 h to appropriate anti-bacterial therapy the probability of both death and neurological impairment were 46.7% (95% CI 27.4-66.1%) and 48.1% (95% CI 26.4-69.6%), respectively. By univariate analysis, increased TTA was associated with increased mortality (OR 1.8 , 95% CI 1.1-2.9) in patients treated after one hour compared to (OR 3.0. 95% CI 2.0-4.5) patients treated after five hours (Table 3).
It was possible to examine the contribution of microbial pathogens on death due to CABM in univariate analysis. S. pneumoniae, S. aureus, and the 'other Gram-negative group' were all associated with a marked increase in the odds of death compared to CABM due to N. meningitidis (Table 3).
Multivariable analysis (excluding the data from Dr Miner et al. as these consisted only of patient age, time to antibiotics and outcomes) showed that each hour delay in time to antibiotics was associated with a 10% increase in odds of mortality (adjusted OR 1.1, 95% CI 1.0-1.1). The adjusted analysis indicated that death was associated with advancing age, reduced conscious state at the time of hospital presentation, and the presence of comorbidities. None of the microbial pathogens was significantly associated with death in the adjusted analysis (Table 4). www.nature.com/scientificreports/  www.nature.com/scientificreports/ Using the individual patient data available, multivariable analysis showed an uncertain impact of TTA on neurological impairment (aOR 1.0, 95% CI 1.0-1.1) ( Table 3). The presence of impaired consciousness at admission and comorbidities remained significantly associated with the likelihood of neurological impairment and, in contrast to the likelihood of death in this analysis, the use of steroids was protective (aOR 0.4, 95% CI 0.3-0.8).
We analysed the relationship of TTA on death and neurological impairment in groups of patients with CABM who were propensity-matched for age and gender. The propensity-matched groups were used to analyse the CABM outcomes in relation to the same discrete hourly intervals to antibiotic administration (less than or greater than one hour up to less than or greater than five hours) as above. The sample sizes for the propensity-matched groups ranged from 180 to 362 patients. Matching for other predictive variables reduced the sample size to the extent that meaningful comparisons were not possible. This analysis gave subtly different results to the above multivariable analysis on the pooled, individual patient data.
A progressive increase in mortality was shown in the propensity-matched groups that was significantly greater for patients treated after each hourly interval (Table 4 and Fig. 3). Notably, while there was a 2% (aOR 1.02, 95% CI 1.02-1.02) increase in odds of mortality if antibiotic therapy was instituted after one hour, this increased to 129% (aOR 2.29, 95% CI 1.28-4.09) where antibiotics were commenced after two hours. The significant association between age, presence of comorbidities, and decreased conscious state at presentation with increased probability of death were of the same magnitude across the periods of one to five hours to anti-bacterial therapy. The use of steroids had a significant protective effect on mortality that was present after a comparison of less than or greater than two hours to antibiotic therapy. After an interval of three hours, time to antibiotics was significantly associated with the probability of neurological impairment (aOR 1.79, 95% CI 1.03-3.14). As with TTA and CABM mortality the presence of impaired consciousness and steroid administration was also significantly associated with the likelihood of neurological impairment (Table 4 and Fig. 3).

Discussion
With the use of multivariable analysis of propensity-matched patients, we have shown that progressive delay between presentation to hospital with microbiologically proven bacterial meningitis and treatment with antibacterial therapy is associated with increased risk of both death and neurological impairment. After a two-hour delay to antibiotic therapy, the odds of death due to CABM is increased 1.3 fold and continues to rise thereafter. Patients treated before or after one hour had almost the same likelihood of death with an adjusted odds ratio of 1.02. We believe that the very narrow confidence intervals for this and other results are related to precision of the estimated effect size. As such, they are credible even though the lower or the upper limit might be equivalent to the effect size. A similar association was present for the probability of neurological impairment measured by the Glasgow Outcome scale on hospital discharge. And for both adverse outcomes, reduced conscious state at hospital presentation, along with the protective effect of steroid administration, were also significant predictive factors.
These results relied on individual patient data from five cohorts of adult cases of CABM. These include new data on adult patients with CABM treated at TUH along with information from Denmark, UK, and North Table 4. Factors associated with mortality and neurological impairment due to community-acquired bacterial meningitis, including time to antibiotics using patients matched for age and sex by propensity scoring. a Less than or greater than hourly interval to the administration of effective antibacterial therapy. b Impaired consciousness at presentation to hospital. www.nature.com/scientificreports/ www.nature.com/scientificreports/ American hospitals. Our attempt to combine all the aggregate data reported in the published literature on this topic has been thwarted due to the marked heterogeneity of study methodologies with incompatible reporting of TTA. The use of individual patient data allowed for improved precision of our estimates of the association between TTA and death and neurological impairment compared to single-institution studies that have been published previously. While CABM occurring in children has decreased in incidence due to the use of conjugate vaccines against S. pneumoniae and N. meningitidis, the burden of disease in adults is increasing. Over the period of 1998-2007 in the USA, the median age of cases of CABM increased from 31 to 42 years and the case fatality rate (~ 16%) remained unaltered 28 . Early initiation of effective anti-bacterial therapy is crucial to the outcome of CABM as the disease progresses rapidly with death and neurological impairment developing as its devastating consequences 2 . The pathogenesis of CABM is complex but death and prolonged neurological impairment result from the severe inflammatory cascade initiated by the virulent causative pathogens 29 . The multiple causes of death include neurological complications such as intracranial bleeding, stroke, status epilepticus, and uncontrolled sepsis 29,30 . Cerebellar tonsillar herniation due to cerebral oedema accounts for 10% of deaths due to CABM 30 . Cerebral arteriopathy due to vasculitis, vasospasm, or intra-arterial thrombosis is the main cause of ischaemia and infarction 31 .
One important caveat associated with this analysis is that overall rates of mortality and neurological impairment are high in comparison to other cohort studies of Western patients 28 . The distribution of causative pathogens from the Danish studies is somewhat atypical in that a significant proportion of cases of adult CABM were due to gram-negative bacteria other than N. meningitidis and H. influenzae, such as Enterobacteriaceae, P. multocida, and Capnocytophaga species. The two Danish studies describe large, community-based samples and they report a very similar distribution of CABM pathogens. CABM due to E. coli is rare and has an extremely high mortality rate 32 . P. multocida 33 and C. canimorsis are also uncommon causes of meningitis occurring after cat and dog contact 34 . P. multocida meningitis has a high mortality while that due to C. canimorsis is low but is a frequent cause of long-term neurological impairment. The other studies included here mainly describe patients with the characteristic CABM pathogens, S. pneumoniae and N. meningitidis.
The main limitation of the study is that while a large number of patient data are available for analysis in this study, they still represent less than a quarter of all the cases described in the literature with a time function related to the commencement of antibiotics. Despite strenuous efforts, it was not possible to obtain other patient data for inclusion. Crucially, due to the observational nature of our data, important confounders may contribute to the measured association between TTA and mortality. For example, earlier antibiotic therapy may be given to patients with more clinically obvious features of CABM and this may correlate with factors associated with good outcomes such as immunological function and lack of frailty, cognitive impairment, or communication difficulty. Indeed, a 2006 systematic review of antibiotic therapy 35 concluded that confounding of observational study data on meningococcal disease obscured the presence of any benefit from early antibiotic administration.
A recent, randomised trial of early antibiotic therapy administered in ambulances for the broader group of infections causing sepsis failed to show reduced mortality compared with routine practice of treatment commencement in emergency departments 36 . The majority of patients studied had severe sepsis due to pneumonia and < 1% (11/2672) of subjects had central nervous system infection so its direct relevance to the treatment of CABM is limited.
Trials of adjuvant corticosteroids for the treatment of adult CABM show reduced mortality is confined to disease due to S. pneumoniae 37 . While these studies do not generally include a time from hospital presentation to antibiotic treatment function it may be that reanalysis of the randomised controlled trial data could provide some additional clarity on the risk of death due to CABM after delayed anti-bacterial therapy.
This study of a large cohort of patients with bacteriologically proven CABM using data from four countries shows that delay of effective treatment for more than two hours is associated with a more than double odds of death. These data could be used to reinforce the appropriate time to antibiotic administration in guidelines for treatment of CABM. Antibiotics must be commenced urgently in patients presenting to our hospitals with clinical features compatible with CABM. This study provides evidence that the time from door to needle must not be more than 2 h. www.nature.com/scientificreports/