Enhanced Influenza Surveillance Using Telephone Triage and Electronic Syndromic Surveillance in the Department of Veterans Affairs, 2011-2015

Abstract

Telephone triage (TT) is a method whereby medical professionals speak by telephone to patients to assess their symptoms or health concerns and offer advice. These services are often administered through an electronic TT system, which guides TT professionals during the encounter through the use of structured protocols and algorithms to help determine the severity of the patients’ health issue and refer them to appropriate care. TT is also an emerging data source for public health surveillance of infectious and noninfectious diseases, including influenza. We calculated Spearman correlation coefficients to compare the weekly number of US Department of Veterans Affairs (VA) TT calls with other conventional influenza measures for the 2011-2012 through 2014-2015 influenza seasons, for which there were a total of 35 666 influenza-coded TT encounters. Influenza-coded calls were strongly correlated with weekly VA influenza-coded hospitalizations (0.85), emergency department visits (0.90), influenza-like illness outpatient visits (0.92), influenza tests performed (0.86), positive influenza tests (0.82), and influenza antiviral prescriptions (0.89). The correlation between VA-TT and Centers for Disease Control and Prevention (CDC) national data for weekly influenza hospitalizations, influenza tests performed, and positive influenza tests was also strong. TT correlates well with VA health care use and CDC data and is a timely data source for monitoring influenza activity.

National experience with pandemic influenza has highlighted the use of syndromic surveillance to track diseases and inform decision making.¹ Ongoing threats from pandemic influenza, as well as other infectious pathogens, reinforce the need for robust and timely systems. With early detection of influenza a priority, nontraditional data sources (ie, absenteeism, over-the-counter drug sales, internet searches, and telephone triage [TT]) are being evaluated as additional sources of surveillance information.^2–14 TT is a method whereby a medical professional, usually a registered nurse, speaks with a patient or caregiver by telephone to assess the patient’s symptoms or health concerns and offer medical advice. TT is an interactive process between the TT professional and the client, with the goal of directing the caller to the appropriate level of care or service in a safe and timely manner. These services are often administered through an electronic TT system, which guides the TT professional during the encounter through the use of structured protocols and algorithms to help determine the severity of the patient’s health issue and the urgency of the patient’s health care needs and then refer the patient, if needed, to the most appropriate care setting.

The US Department of Veterans Affairs (VA) is an integrated health care system with approximately 9 million enrollees who are served at more than 1200 sites throughout the United States and US territories. Nearly 46% of patients are aged ≥65, and a small but growing percentage (9%) is female, making the VA population unique compared with other populations under national surveillance.¹⁵ The Electronic Surveillance System for the Early Notification of Community-based Epidemics (ESSENCE), the syndromic surveillance tool used by the VA from 2005 to 2015, originally focused on syndrome categories based on data from outpatient encounters.¹⁶ In 2009, VA-ESSENCE incorporated inpatient and TT data and was used to track influenza activity in the VA during the H1N1 influenza pandemic and also to evaluate influenza-like illness (ILI) trends in combined VA and US Department of Defense patient populations.^17,18 In 2015, the VA transitioned to another syndromic surveillance tool called Praedico (Bitscopic, Inc), which has enhanced architecture and flexibility to accommodate the VA’s growing volume and variety of data.^19,20

Influenza causes substantial morbidity and mortality in VA patients, although the burden of disease fluctuates from year to year. For the 6 most recent influenza seasons (2010-2011 through 2015-2016), activity was lowest during the 2011-2012 influenza season: 546 influenza-coded hospitalizations and 1005 laboratory-confirmed cases recorded. Activity was highest during the 2014-2015 influenza season: 4673 hospitalizations and 11 506 laboratory-confirmed cases. However, hospital deaths for these same seasons were consistent at 3% each season.²¹

Since 2001, the VA has provided TT services through Veterans Health Gateway (VHG) (DSHI Systems, Inc), which is a web-based, standardized triage decision-making solution used by thousands of VA-TT nurses.²² Using structured protocols and proprietary algorithms, nurses use the VHG clinical software application to systematically assess the patient’s chief complaint, positive and negative symptom responses, available values and measures (eg, temperature), and medical history. The system provides TT nurses with clinical decision support and makes recommendations on the most likely diagnosis and where and how quickly the patient should be seen by a provider (eg, emergency department for serious problems or outpatient clinic for less urgent problems). It also identifies minor illnesses that can be managed at home and provides self-care or interim-care instructions. VHG clinical records are entered in the patient’s electronic medical record as a triage encounter record, and all records undergo physician review to ensure quality control.²²

VA-TT has many desirable characteristics for syndromic surveillance. It collects information for mild illnesses among patients who never seek in-person medical attention. For patients who do seek in-person medical attention, TT occurs before they access other health care resources (eg, office visits). Importantly, TT collects data on demographic characteristics, chief complaints, coded descriptions of medical problems, and recommended follow-up. TT data are also desirable because they are available in near-real time, providing an advantage over other data sources, such as hospitalization records, which are not available until after a patient is discharged and the record is closed. This can take ≥1 week after the discharge date. To systematically evaluate the use and timeliness of TT, we compared VA-TT influenza–related calls with other influenza metrics (outpatient encounters, hospitalizations, ILI outpatient visits, influenza testing, and antiviral prescriptions) using previously described methods, including correlation, peak comparison, and aberration detection comparison.²³

Methods

Study Population, Data, and Definitions

We analyzed all influenza-specific International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM)²⁴ outpatient encounters, hospitalizations, TT calls, ILI outpatient visits, influenza testing, and antiviral prescriptions for October 2, 2011, through September 26, 2015, which covered 4 influenza seasons (2011-2012 through 2014-2015) for all VA sites of care. These VA data have previously been evaluated and are routinely used for influenza surveillance.^17–19,21 The surveillance data used in this study were acquired as part of VA operations and public health activities. As such, VA or facility institutional review board review was not required.

Outpatient encounters are grouped into the ILI syndrome if assigned ICD-9-CM code(s) match a designated ILI syndrome code (Box). TT call time is recorded, along with data on demographic characteristics of the patient, chief complaint, recommended follow-up timing and location, best-fit ICD-9-CM diagnosis code(s), and system concern. The system concern is the condition or brief list of conditions that justify the recommended follow-up and represent the most concerning condition(s) that could account for the patient’s symptoms; however, it might not represent the actual diagnosis.

Box.

International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM)²⁴ codes and code combinations included in the US Department of Veterans Affairs influenza-like illness syndrome category

Code	Description
079.89	Other specified viral infection
079.99	Unspecified viral infection
460	Acute nasopharyngitis (common cold)
462	Acute pharyngitis
465.8	Acute upper respiratory infections of other multiple sites
465.9	Acute upper respiratory infections of unspecified site
466.0	Acute bronchitis
480.9	Viral pneumonia, unspecified
484.8	Pneumonia in other infectious diseases classified elsewhere
485	Bronchopneumonia, organism unspecified
486	Pneumonia, organism unspecified
487.0	Influenza with pneumonia
487.1	Influenza with other respiratory manifestations
487.8	Influenza with other manifestations
488.01	Influenza due to identified avian influenza virus with pneumonia
488.02	Influenza due to identified avian influenza virus with other respiratory manifestations
488.09	Influenza due to identified avian influenza virus with other manifestations
488.11	Influenza due to identified 2009 H1N1 influenza virus with pneumonia
488.12	Influenza due to identified 2009 H1N1 influenza virus with other respiratory manifestations
488.19	Influenza due to identified 2009 H1N1 influenza virus with other manifestations
488.81	Influenza due to identified novel influenza A virus with pneumonia
488.82	Influenza due to identified novel influenza A virus with other respiratory manifestations
488.89	Influenza due to identified novel influenza A virus with other manifestations
490	Bronchitis, not specified as acute or chronic
780.60 + 784.1	Fever, unspecified plus throat pain
780.60 + 786.2	Fever, unspecified plus cough

VA-ESSENCE uses aberrancy-detection algorithms employing regression modeling, as previously described, and produces alerts when observed counts are significantly above the model predictions.^16,25,26 VA encounter and TT data for the 2011-2012 through 2013-2014 influenza seasons came from VA-ESSENCE. Data for the 2014-2015 influenza season came from VA-ESSENCE and Praedico. We only used VA-ESSENCE algorithms in this analysis. Influenza testing and antiviral data were extracted from the laboratory and pharmacy domains within VA data warehouses, which are national repositories of data from VistA, VA’s electronic medical record system, and other VA clinical and administrative systems. For external comparison, we used Centers for Disease Control and Prevention (CDC) influenza hospitalization data (FluSurv-NET)^27,28 and laboratory data (US World Health Organization and National Respiratory and Enteric Virus Surveillance System collaborating laboratories)^28,29 from the same seasons, which represent a different patient population from VA.

Analysis

We calculated Spearman rank-order coefficients, 95% confidence intervals (CIs), and P values using the Fisher z transformation to describe the correlation between the weekly number of VA-TT influenza encounters and other weekly influenza measures, as well as the correlation in influenza hospitalization and laboratory data between the VA and CDC. For example, we compared weekly VA-TT encounters (x) with weekly VA influenza hospitalizations (y) to obtain the Spearman rank-order correlation coefficient between x and y. We calculated peaks for health care use during each season for each data source. We plotted VA influenza-coded hospitalizations, deaths, and ILI outpatient visits against TT encounters (the 2012-2013 through 2014-2015 seasons only) to compare alerting trends. ESSENCE system algorithms identified low-level (yellow) warnings for influenza activity when the observed count fell between the 95% and 99% CIs (P < .05) of the expected count. A high-level (red) alert was generated when the value exceeded the 99% CI (P < .01). We did not include alerting for the 2011-2012 season because limited historical data were available to accurately predict aberrancy. We compared the timing of peaks and alerts to assess maximum health care use and the timeliness of surveillance alerting using Welch’s t test for unequal variance. We analyzed data using SAS version 9.3.³⁰

Results

During the study period, there were 35 666 influenza-coded VA-TT calls (33 254 unique callers, representing 1.5% of all VA-TT calls), 23 475 positive influenza tests (of 160 949 tests performed), 18 160 influenza-coded emergency department visits, 9488 influenza-coded hospitalizations, 74 139 influenza antiviral prescriptions, and 1 827 824 ILI outpatient encounters. Of the 33 254 unique callers, 27 501 (82.7%) were male and 23 029 (69.3%) were aged <65.

The weekly number of influenza-coded calls was strongly correlated with weekly influenza-coded hospitalizations (r = 0.85; 95% CI, 0.80-0.88), emergency department influenza-coded visits (r = 0.90; 95% CI, 0.87-0.92), ILI outpatient visits (r = 0.92; 95% CI, 0.90-0.94), influenza tests performed (r = 0.86; 95% CI, 0.82-0.89), positive influenza tests (r = 0.82; 95% CI, 0.77-0.86), and influenza antiviral prescriptions (r = 0.89; 95% CI, 0.86-0.91) (P < .001 for all correlations). Correlation coefficients for calls with an influenza-related system concern (conditions defined as influenza, mild influenza, or severe influenza) were similar to results for influenza-coded calls. Weekly VA-TT and CDC data for influenza-coded hospitalizations, influenza tests performed, and positive influenza tests were also strongly correlated (P < .001 for all correlations). We did not compare VA-TT counts with CDC ILI counts because of variations in the number and type of ILINet reporting providers by week and season²⁹ (Table).

Table.

Spearman rank-order correlation coefficients comparing weekly VA telephone triage influenza encounters with weekly VA and CDC health care use data, October 2, 2011, to September 26, 2015

Telephone Triage Data	Health Care Use	Correlation Estimate (95% CI)a
Calls with ICD-9-CM 24 diagnosis code of influenza (N = 35 666)	VA influenza-coded hospitalizations (n = 9488)	0.85 (0.80-0.88)
	VA emergency department influenza-coded visits (n = 18 160)	0.90 (0.87-0.92)
	VA ILI outpatient visits (n = 1 827 824)	0.92 (0.90-0.94)
	VA influenza tests performed (n = 160 949)	0.86 (0.82-0.89)
	VA positive influenza tests (n = 23 475)	0.82 (0.77-0.86)
	VA influenza antiviral prescriptionsb (n = 74 139)	0.89 (0.86-0.91)
	CDC FluSurv-NET hospitalizations (n = 41 872)	0.82 (0.75-0.87)
	CDC FluView influenza tests performed (n = 1 835 079)	0.84 (0.79-0.88)
	CDC FluView positive influenza tests (n = 294 896)	0.82 (0.78-0.87)
Calls with system concern for influenzac (N = 25 443)	VA influenza-coded hospitalizations (n = 9488)	0.85 (0.81-0.89)
	VA emergency department influenza-coded visits (n = 18 160)	0.83 (0.79-0.87)
	VA ILI outpatient visits (n = 1 827 824)	0.80 (0.74-0.84)
	VA influenza tests performed (n = 160 949)	0.92 (0.90-0.94)
	VA positive influenza tests (n = 23 475)	0.86 (0.82-0.89)
	VA influenza antiviral prescriptionsb (n = 74 139)	0.89 (0.86-0.91)
	CDC FluSurv-NET hospitalizations (n = 41 872)	0.84 (0.77-0.88)
	CDC FluView influenza tests performed (n = 1 835 079)	0.89 (0.86-0.92)
	CDC FluView positive influenza tests (n = 294 896)	0.81 (0.76-0.85)

Abbreviations: CDC, Centers for Disease Control and Prevention; ICD-9-CM, International Classification of Diseases, Ninth Revision, Clinical Modification; ILI, influenza-like illness; VA, US Department of Veterans Affairs.

^a P < .001 for all correlation coefficients.

^bIncludes inpatient and outpatient orders for oseltamivir, zanamivir, and peramivir.

^cSystem concern for influenza includes triage differential diagnoses of influenza, mild influenza, or severe influenza.

Each season, influenza-coded telephone calls peaked during the same week or within 1 week of influenza-coded hospitalizations, outpatient and emergency department influenza-coded visits, percentage of patient visits for ILI, influenza tests performed, positive influenza tests, and influenza antiviral prescriptions, except for during the 2011-2012 season, during which ILI outpatient visits peaked earlier. System alerting for increased activity began first with TT data for each season evaluated compared with other VA influenza measures (Figure). Initial TT alerts occurred on average 11 weeks before alerts for influenza hospitalizations (P = .001) or deaths (P = .04) and 16 weeks before alerts for percentage of outpatient visits for ILI (P < .001).

Figure.

US Department of Veterans Affairs (VA) telephone triage (TT) encounter data for influenza-coded calls, hospitalizations, and associated deaths and outpatient influenza-like illness (ILI) visits for the 2011-2012 through 2014-2015 influenza seasons, United States. Alerts were limited to the 2012-2013 through 2014-2015 influenza seasons because of a lack of historical data for accurate aberrancy detection in 2011-2012.

Discussion

VA-TT is a promising data source for syndromic surveillance. It is timely and captures a wide spectrum of influenza illness severity. For mild presentations, TT may represent the only interaction between patients and the VA health care system. VA-TT calls are strongly correlated with influenza outpatient and ILI visits, influenza hospitalizations, influenza antiviral prescriptions, and influenza testing. Increases in VA-TT calls occurred before ILI visits and influenza-coded hospitalizations. TT was the only data source that provided routine early season alerting, although peak activity was generally identified during the same week for all data sources. This finding suggests that TT data could augment VA influenza surveillance efforts.

This study is the first to evaluate the use of TT data for influenza surveillance within VA. We found that VA-TT correlates well with traditional VA influenza surveillance measures and with national CDC hospitalization and influenza testing data. Our findings confirm previous evaluations that found moderate to high correlation and timeliness of TT data compared with ILI surveillance and international health agency–reported or laboratory-confirmed cases of influenza.^4,7,8,12,13 In 1 study, national triage call data from 17 states had a median correlation (using the Pearson correlation test) of 0.65 between call data and CDC viral isolate data,⁴ whereas the correlation of VA-TT calls with CDC viral isolate data in our study was 0.82. Another study from Ireland found correlations between influenza-related TT calls and national ILI rates as high as 0.90.¹³ The correlation of VA-TT calls and VA ILI visits in our study was 0.92. National surveillance systems in the United Kingdom^10,12 as well as regional systems in Canada,^6,31 Ireland,¹³ and Switzerland¹⁴ have successfully incorporated TT into their routine influenza surveillance. Although demonstrations of regional TT data in the United States have been analyzed as a possible adjunct to local- or state-level surveillance,^4,7 currently no national system in the United States incorporates TT to monitor influenza. The VA could provide a robust early warning system for influenza if existing surveillance was combined with TT data.

One major challenge in using TT data for influenza surveillance is accurately distinguishing influenza from other circulating respiratory viruses. An influenza diagnosis in VA-TT is assigned by the system algorithms based solely on the brief history provided, symptom responses, and possibly some self-reported measures (eg, temperature). It may be more difficult to distinguish influenza from other respiratory viruses in this setting than in a face-to-face encounter, in which a physical examination can be performed and laboratory testing and/or radiology studies can be ordered. Other researchers have found sizable contributions and confounding effects of other pathogens (particularly respiratory syncytial virus, Streptococcus pneumoniae, and rhinovirus) for TT calls related to colds and influenza.^11,12 Further research is needed to determine whether our observed early TT alerting reflects false alerting, increases in other seasonal respiratory diseases, or a true early warning for influenza.

Limitations

This study had several limitations. First, the variation in telephone call volume across VA geographic regions was substantial. Although VA-TT is available in all VA regions, it is not used uniformly across the VA because at some facilities, patients have an option to telephone a clinic directly to schedule an office visit or make a same-day appointment. In addition, administrative data may not accurately capture the true burden of influenza because of ICD-9-CM coding irregularities, including undercoding, overcoding, and miscoding of influenza.³² TT may be alerting increased activity before other influenza measures because of signals from other respiratory illnesses. One advantage of the VA’s recent transition to Praedico is improved specificity and precision for detecting syndrome clusters and reduced alerting.²⁰ Moving forward, we can evaluate whether timing and the number of TT alerts are improved with this system. Finally, we performed our analysis against a single external data source (CDC FluView). Additional external comparisons would provide a more robust analysis of VA-TT data as a potential early warning system for influenza outbreaks.

Conclusions

VA-TT data are routinely included in VA influenza surveillance reports, which are provided to decision makers at the VA and CDC. VA-TT data could provide valuable information for public health action during another pandemic or during a vaccine or antiviral shortage. Having a tool that incorporates laboratory, pharmacy, encounter, procedural, and TT data in a single application with the ability to examine detailed information by patient, date, and geography using linked data sources will enhance validity and specificity and could provide earlier predictive capability. TT also holds promise for situational awareness of gastrointestinal outbreaks⁹ and other syndromes, which can be explored using VA-TT data.³³ Additional analyses are needed to correlate TT with health care use and illness severity. As enhancements to VA surveillance data are implemented, the VA can investigate geographic disparities or other reasons for those patients who report directly to facilities without using TT to refine how TT can better inform surveillance efforts.