Predictors of influenza among older adults in the emergency department

We prospectively recruited community-dwelling older adults age ≥60 years presenting to the EDs of six acute care hospitals in urban and suburban areas of south-central Ontario, Canada. These EDs had a median of 175 visits per day (range 160–350). Patients or their substitute decision-makers were approached for consent, nasopharyngeal swabs were collected, and, whenever possible, interviews were conducted while they were in the ED.

Participant enrollment occurred during weeks when the proportion of specimens submitted to Ontario laboratories testing positive for influenza was greater than 5 % for the 2011/12 and 2012/13 seasons (Fig. 1). Study personnel screened and enrolled patients from each site during five, 8-h shifts per week. The shifts were scheduled to sample patients in each ED on all seven days of the week for 24 h of the day. Study eligible patients who had most recently arrived were approached first by the research assistant to avoid bias introduced by recruiting patients with longer ED stays. Study approval was obtained from all participating hospital research ethics boards (REBs).

Eligible patients had chief complaints of cardiac, respiratory or cerebrovascular events, systemic symptoms (fever, weakness) or altered level of consciousness. Patients with chief complaints related to trauma, mental disorders, post-operation complications and/or urogenital problems were not eligible. For other chief complaints, patients were included if their triage record indicated any of the following symptoms or conditions: respiratory symptoms, triage temperature ≥37.5 °C, feverishness, malaise, or general weakness. Residents of long-term care homes or other group settings (e.g., detention centers) and patients who had been enrolled in the study within the previous 14 days were not eligible.

Recruiters interviewed participants to determine symptoms, lifestyle, healthcare use, history of pneumococcal vaccination, and influenza vaccination in the previous five years, medical care/consultation during current illness episode, and Clinical Frailty Scale items [15]. Demographic, medical history, and symptom information were also collected by chart review. The participant’s vaccine provider and/or family physician was contacted to obtain influenza vaccination history (for the present and previous season), pneumococcal vaccination history and information on prior visits related to the current episode of illness. Vaccine providers were re-contacted to confirm vaccination information if provider and patient-reported status were discrepant. If the discrepancy could not be resolved, the ‘yes’ response from either source was accepted. Similarly, for discrepancies between self-reported and chart data for symptoms or medical history, a ‘yes’ from either source was taken as the final response.

The outcome of interest was laboratory-confirmed influenza infection. A nasopharyngeal swab collected on date of enrolment was tested for influenza by reverse transcription PCR (RT-PCR) at a single study laboratory within 24 h of collection using bioMérieux easyMAG with RealStar® influenza RT-PCR (Altona Diagnostics) in 2011/12 and Simplexa™ Flu A/B & RSV Direct RT-PCR (Focus Diagnostics) in 2012/13.

We chose definitions of predictors to maximize sensitivity and best discriminate between those with and without influenza, with preference given to variables being readily available in ED settings. We assessed self-reported and chart-recorded severity (mild, moderate or severe) of clinical symptoms as well.

The definition of cough was a composite of self-reported and/or chart-recorded data. Participants were recorded as having cough if it was reported during the interview as a symptom prior to the ED visit or if cough was recorded in the medical chart. Fever was defined as self-reported feverishness and/or having a triage-recorded temperature ≥37.2 °C [13]. All participants were asked if they felt feverishness during their current episode of illness. Acute onset of symptoms was defined as being sudden (i.e., woke up feeling miserable or were unable to do regular activities within a few hours) versus gradual development (i.e., symptoms worsened over several days). Time from symptom onset to ED visit was calculated as the number of days between participant-reported day of onset and presentation to the ED.

A participant was defined as having received the influenza vaccine if it was received in the current season and ≥14 days before symptom onset. Participants' chronic conditions were abstracted from medical records and medication history was collected by interview and medical chart review.

Participants were asked if they lived with children (≤16 years), and whether they had regular contact (≥4 h per day and for ≥4 days per week) with other children outside the household. Recent use of public transportation included riding a bus, subway, train, or streetcar within four days of symptom onset. Exposure to a person with an ILI was defined as being within arm’s length for ≥2 min to someone the participant recognized as having symptoms of an acute respiratory illness (i.e., had cough, stuffy or runny nose, hoarseness or a sore throat) within 7 days of symptom onset.

Influenza activity in the community was defined as the proportion of specimens positive for influenza submitted to laboratories in south-central Ontario during the past week. Neighborhood-based material and social deprivation indices were determined by matching participants’ postal codes with the Canadian Deprivation Index database [16]. The index is divided into 5 quintiles with ‘1’ being the most privileged and ‘5’ representing the most deprived.

Ethnicity was self-reported by participants. Frailty was defined as having a score of ≥5 on the Clinical Frailty Scale (maximum 9 points), as scored by a trained research assistant [15], where the participant required at least some assistance with daily activities [17].

We compared the characteristics of individuals having laboratory-confirmed influenza with those who tested negative. Chi-square and Fisher’s exact tests were applied to assess differences among group proportions. Medians were compared using Wilcoxon rank-sum tests. Reported p-values are two-sided with a value <0.05 considered to be statistically significant. No correction was made for multiple testing. Odds ratios (OR) and 95 % confidence intervals (CI) were estimated using logistic regression. Hypothesized predictors significant at p <0.2 in the bivariate analysis were included in the initial logistic regression model [18]. The full model was reduced by sequentially removing predictors that did not significantly improve the fit of the model as determined by the Likelihood Ratio Test [18]. Hosmer-Lemeshow Goodness-of-Fit Test and residuals plots were used to assess model fit. Variance inflation factors (VIF > 10) were used to assess for collinearity between predictors. No imputation was done for missing data. In the multivariable analysis, participants with missing data on included predictors were excluded. The final model was adjusted for hospital site and had at least ten events per predictor in the model.

Statistical interaction terms, determined a priori, were entered as cross-product terms in a logistic regression model. If significant in the crude analysis, the interaction term was added to the final model and assessed for significance [19]. To interpret effect modification, the odds ratio was calculated for one predictor on the outcome over the levels of another predictor. Vaccine effectiveness was calculated as (1 - OR) X 100.

To explore potential differences between influenza types (A & B), we compared each type to influenza-negative participants. We used SAS version 9.3 (SAS Institute, Cary, NC, USA) for all analyses.