Bulletin of the World Health Organization

Performance of case definitions used for influenza surveillance among hospitalized patients in a rural area of India

Siddhivinayak Hirve a, Mandeep Chadha b, Pallavi Lele a, Kathryn E Lafond c, Avinash Deoshatwar b, Somnath Sambhudas a, Sanjay Juvekar a, Anthony Mounts c, Fatimah Dawood c, Renu Lal c & Akhilesh Mishra b

a. KEM Hospital Research Centre, Sardar Moodliar Road, Rasta Peth, Pune, 411011, Maharashtra, India.
b. National Institute of Virology, Pune, India.
c. Centers for Disease Control and Prevention, Atlanta, United States of America (USA).

Correspondence to Siddhivinayak Hirve (e-mail: sidbela@vsnl.com).

(Submitted: 14 June 2012 – Revised version received: 15 August 2012 – Accepted: 16 August 2012 – Published online: 10 October 2012.)

Bulletin of the World Health Organization 2012;90:804-812. doi: 10.2471/BLT.12.108837

Introduction

Case definitions for influenza surveillance need to be simple, easily understood and easily implemented. The international standardization of such case definitions is difficult, partly because the optimal choice of case definition depends on the population involved and the objectives of the surveillance. The World Health Organization (WHO) arranged a consultation on this topic in 2011 and is currently developing a document to describe minimum global standards for influenza surveillance case definitions. In the surveillance of severe influenza requiring hospitalization, the WHO currently recommends a particular case definition for severe acute respiratory illness: “cough or sore throat, plus measured fever, shortness of breath and need for hospitalization” (Table 1). In the surveillance of outpatient influenza, however, WHO recommends using the case definition for influenza-like illness (measured fever plus cough or sore throat). Despite this recommendation, in some countries a case definition for acute respiratory illness is used to evaluate outpatient influenza.

The case definitions used in national and regional influenza surveillance programmes often differ from those recommended by WHO.18 However, most case definitions used for influenza surveillance include measured or reported fever as well as cough and/or sore throat. Some also include constitutional symptoms, such as joint pain, myalgia, headache and/or malaise. A case definition for febrile acute respiratory illness has been used in the detection of influenza in research settings.9

The performance of influenza case definitions has been evaluated many times but only among ambulatory or hospitalized patients with fever or respiratory symptoms.1020 Such evaluations may have missed influenza cases with atypical clinical presentations, particularly those caused by acute exacerbations of underlying disease. In the Hong Kong Special Administrative Region of China and in the United States of America, > 75% of all cases of influenza-associated severe disease have been found to be classified not as influenza but under the underlying condition (e.g. cardiovascular disease, chronic lung disease or diabetes) that placed the individuals at risk.21,22

We conducted a study to estimate the burden of hospitalized influenza in a rural community in western India between 2009 and 2011. To maximize the detection of patients with hospitalized influenza, we screened all hospital admissions and enrolled patients with any acute medical illness of recent onset, including acute exacerbations of underlying chronic conditions. After using a reverse-transcription polymerase chain reaction (PCR) to identify patients infected with influenza virus, we retrospectively evaluated the sensitivity and specificity of various case definitions, including those that are commonly used for the detection of influenza among hospitalized patients.

Methods

Setting

The study area of Vadu lies in Pune district, 30 km to the north-east of Pune city, in western India. Vadu has a health-care infrastructure that is typical of a rural area in close proximity to a larger town or city in India. The public sector includes a rural hospital and primary health centres and the private sector includes several small general hospitals. Medical care is easily accessible. Most residents seek care in the private sector, which is largely unregulated. Hospital admission, treatment and pricing guidelines are often unclear, not standardized and subject to the providers’ preferences. For the present study, surveillance for hospitalized influenza cases was conducted in 29 general public and private hospitals (with a mean of 15 beds per hospital) in and around the 22 villages of the Vadu Demographic Surveillance Area (DSA) in Pune district. In the study area, the monsoon and winter seasons run from June to September and November to February, respectively. Pune district has seasonal malaria and dengue activity. Although influenza activity peaks during the monsoon season, some circulation of influenza viruses occurs year-round.23,24

Patient screening and enrolment

Field-based investigators (trained non-clinical research staff) visited each study hospital every day between 1 May 2009 and 30 April 2011 and screened all patients who had been admitted in the previous 24 hours. Patients were excluded if they resided outside the Vadu DSA, if they had been admitted for trauma, elective surgery or obstetric or gynaecological conditions, and/or if they had not spent at least one night in a study hospital. After informed consent was obtained from the adults and the legal guardians of the children, all other patients admitted for any acute medical condition were enrolled, by the study physician, within 24 hours of admission. Those enrolled included patients admitted for onset of respiratory symptoms within the previous week or for acute onset of fever or history of fever within the previous week, and patients admitted for acute exacerbations of pre-existing chronic medical conditions (e.g. chronic lung disease, asthma, cardiovascular disease, stroke or diabetes). Detailed clinical information on each enrolled patient was abstracted from medical records at the time of admission and was also collected by the study physician at the time of interview.

Laboratory methods

A nasal and/or throat swab was collected from each enrolled patient. Within 2 or 3 days, the swabs from each patient were transported to the laboratories of the National Institute of Virology in Pune, at 2–8 °C, in a single vial of virus transport medium.25 Ribonucleic acid (RNA) was then extracted from a subsample (50 µl) of each specimen by using a Magmax-96 viral isolation kit (Ambion®, Life Technologies Corporation, Paisley, Scotland) according to the manufacturer’s instructions. Two real-time PCRs were subsequently used, first to check the RNA for evidence of influenza A and B viruses and then, if the first PCR gave a positive result for influenza A virus, to identify the subtype of influenza A virus present in the specimen.26

Ethics

The study protocol was approved by the Ethics Review Committee of the KEM Hospital Research Centre (Pune, India), the Ethics Review Committee of the National Institute of Virology (Pune, India) and the Institutional Review Board of the Centers for Disease Control and Prevention (Atlanta, USA). Written informed consent was obtained from the adult patients who were investigated, the legal guardians of the children who were investigated and the physicians who participated in the study.

Statistical methods

The proportional distributions of the patients who were found PCR-positive for an influenza virus, split by sociodemographic status and other characteristics, were tested for heterogeneity using Fischer’s exact test. The results of the PCR were used as the “gold standard” in evaluating the performance of various standard case definitions (Table 1), individual symptoms and symptom combinations in the identification of influenza. A risk-factor analysis was conducted to estimate the predictive value of various clinical symptoms in the identification of influenza. Sensitivities, specificities and (for the period when > 10% of enrolled patients were found PCR-positive for an influenza virus) positive and negative predictive values were estimated. Case definitions that included fever were evaluated by limiting the definition of fever to measured fever at the time of presentation to hospital or by expanding the definition of fever to include both measured fever and reported history of fever.

Results

Between May 2009 and April 2011, 10 080 patients were admitted to the study hospitals. Although 9426 (94%) of these patients were screened for residential and clinical eligibility, only 3391 (36% of those screened) met the inclusion criteria and were enrolled and tested for influenza. Of those enrolled, 3179 were included in the final analysis; the other 212 were excluded because of inadequate specimen quality (n = 55) or missing clinical information (n = 157). Overall, 665 (21%) of the patients included in the analysis were found PCR-positive for an influenza virus: 20 (3% of the PCR-positives) for seasonal A(H1N1), 109 (16%) for seasonal A(H3N2), 340 (51%) for pandemic A(H1N1)pdm09 and 196 (30%) for influenza B virus. Nasal and throat swabs were collected from most (76%) of the tested patients 3–7 days after the onset of symptoms. The other swabs were collected < 3 (13%) or > 7 days (11%) after symptom onset.

The proportion of admissions found PCR-positive for an influenza virus was significantly higher in the first year of the study period (i.e. May 2009–April 2010) than in the second year (i.e. May 2010–April 2011): 23% versus 19% (P = 0.007). Among PCR-positive patients, pandemic A(H1N1)pdm09 virus predominated in both the first and second years (46% and 56%, respectively), followed by the influenza B (19% and 40%, respectively) and seasonal influenza A(H3N2) viruses (29% and 4%, respectively). Most (59%) of the PCR-positive patients were aged 5–29 years but 22% were children aged < 5 years (Table 2). The positivity rate for pandemic A(H1N1)pdm09 was not significantly different from that for the seasonal influenza viruses.

Ninety (3.1%) patients had at least one underlying chronic condition: 87 (3%) were aged ≥ 5 years and three (< 1%) were aged < 5 years. Among these 90 patients with underlying chronic conditions, PCR positivity was most common among those with asthma (26%) or chronic obstructive pulmonary disease (13%); none of the 26 patients with underlying diabetes was PCR-positive.

Respiratory specimens from patients aged ≥ 5 years were more likely to be found PCR-positive if they had been collected within 1 week of symptom onset than if they had been collected longer after symptom onset: 22% versus 17% (P = 0.039). However, the corresponding values for the younger patients were not significantly different (P = 0.296).

Clinical signs and symptoms

Of the 337 patients aged < 5 years, 44 (13%) were found to be PCR-positive for an influenza virus. The proportion of these 337 patients who had information missing for any clinical symptom (e.g. history of fever, cough and shortness of breath) or sign (temperature or respiratory rate) was < 2%. For 5% of these children, however, no attempt was made to record any of the danger signs highlighted by WHO’s guidelines for the Integrated Management of Childhood Illness (IMCI; i.e. inability to drink or breastfeed, vomiting of everything, convulsions, lethargy/unconsciousness, chest in-drawing or stridor in a calm child). Most (93%) of the 337 patients aged < 5 years reported fever and 14% had fast breathing, but only 17% were recorded as having at least one of the IMCI danger signs. Discharge diagnosis data were available for 84% of the patients aged < 5 years; the most common discharge diagnoses for this age group were fever (80%) and pneumonia (8%). Among the patients aged < 5 years, those found PCR-positive were no more (or less) likely to have respiratory symptoms or at least one IMCI danger sign than those found PCR-negative. Those found PCR-negative were, however, more likely to suffer from vomiting (15.3% versus 7.5%; P = 0.047) or diarrhoea (15.8% versus 5.8%; P = 0.011) than their PCR-positive counterparts (Table 3).

Among the 2842 patients aged ≥ 5 years, 621 (22%) were found PCR-positive for an influenza virus. Almost all (96%) of these patients reported fever and 3% had shortness of breath. Cough, sore throat, nasal discharge and earache were more common in the PCR-positive patients of this age group than among the PCR-negative (Table 3), but headache was rarer among the PCR-positive. Discharge diagnosis data were available for 75% of the patients aged ≥ 5 years; the most common discharge diagnoses for this age group were viral fever (76%), malaria (7%) and enteric fever (7%).

Case definition performance

In children aged < 5 years

Of all the case definitions evaluated, that for severe acute respiratory illness gave the lowest sensitivity (11%) but also the highest specificity (87%) among the patients aged < 5 years (Table 4). When shortness of breath was excluded from this case definition (leaving “influenza-like illness with measured fever in a hospitalized patient”), the sensitivity increased to 69% while the specificity decreased to 43%. A further modification – the inclusion of reported or measured fever instead of just measured fever – resulted in another increase in sensitivity (to 87%) and another decrease in specificity (to 5%).

On its own, fever (measured or reported) appeared highly sensitive (95%) but unspecific (3%) in the detection of influenza among hospitalized children aged < 5 years. In contrast, shortness of breath appeared highly specific (86%) but had low sensitivity (11%). The combination of fever (reported or measured) plus cough gave high sensitivity (77%) and moderate specificity (33%). The corresponding positive predictive values ranged from 11–17% for the standard case definitions and 14–16% for individual signs/symptoms or sign/symptom combinations. The corresponding negative predictive values were all > 70%.

In patients aged ≥ 5 years

Among the patients aged ≥ 5 years, case definition for severe acute respiratory illness again had low sensitivity (3%) but high specificity (98%). When shortness of breath was excluded from this definition, sensitivity increased to 70% but specificity decreased to 53%. The case definition for acute respiratory illness gave even better sensitivity (87%) but lower specificity (36%). The addition of measured fever to this case definition lowered sensitivity, whereas the addition of reported fever had no effect on sensitivity or specificity. On their own, fever and shortness of breath performed similarly among patients aged ≥ 5 years as among the younger patients. Two constitutional symptoms, headache and myalgia, each had high sensitivity (86% and 73%, respectively) but low specificity (10% and 27%, respectively). The combination of measured fever with either cough or sore throat (i.e. influenza-like illness with measured fever) gave high sensitivity (70%) with moderate specificity (53%). Including a constitutional symptom (myalgia or headache) in the case definition of influenza-like illness decreased sensitivity (from 70% to 63%) but slightly increased specificity (from 53% to 56%). The corresponding positive predictive values for the standard case definitions (26–39%) were similar to those calculated for individual signs/symptoms and symptom combinations. As in the younger patients, all of the corresponding negative predictive values exceeded 70%.

Discussion

In the present study, the sensitivity and specificity of various case definitions for influenza were evaluated among > 3000 patients hospitalized with acute medical illness in hospitals in rural western India. Very few published studies have evaluated influenza case definitions among hospitalized patients with a similarly broad spectrum of acute illness, including non-respiratory illness.17,27 Although WHO’s case definition for severe acute respiratory illness has frequently been used to screen for influenza among hospitalized patients, our findings indicate that the use of this case definition (both among hospitalized children aged < 5 years and among older hospitalized patients) results in a low detection of influenza cases, although those meeting the case definition are likely to harbour an influenza virus. Similarly low sensitivities have been recorded for this case definition in other studies, at least among patients aged ≥ 5 years.27 Since, in the present study, patients were checked for shortness of breath, difficulty breathing and tachypnea by trained study physicians, the low numbers of patients with these problems were unlikely to be attributable to under-detection. These problems may have been encountered relatively rarely in the present study because concerns raised by the influenza pandemic that occurred during the study period (i.e. the 2009 H1N1 pandemic) encouraged the patients investigated to seek medical care relatively early in their illness. The exclusion of shortness of breath from the case definition for severe acute respiratory illness (leaving a definition equivalent to influenza-like illness in a hospitalized patient) or the use of a simpler definition of “fever plus cough” improved sensitivity without substantial loss of specificity. Such simplified case definitions are also easier to apply in a surveillance setting because they require screening for fewer criteria. Broadening the case definition for influenza-like illness or the “fever plus cough” case definition, to include reported fever as an alternative to measured fever, further increased sensitivity while still keeping the screening criteria fairly simple.

Influenza case definitions have been evaluated among hospitalized patients before, but most relevant studies were, until recently, largely limited to adults hospitalized in developed countries, where the spectrum of underlying illness and access to care are, in general, substantially different from those seen in developing regions.1517,19 The present study, in a rural area of India, benefitted from the high volume of admissions to the study hospitals and the high prevalence of influenza in the study population. These features permitted the performances of various case definitions for influenza to be evaluated separately among hospitalized patients belonging to two age groups (i.e. children aged < 5 years and older individuals). As previously reported,28 the clinical presentations of the hospitalized children aged < 5 years with and without influenza were similar, such that most case definitions for influenza in this age group had low specificities. Unfortunately, the signs and symptoms of influenza show substantial overlap with those of the infection with other respiratory viruses, such as respiratory syncytial virus, that account for a large proportion of hospitalizations among children aged < 5 years.29 Since the performances of all the case definitions commonly used to screen for influenza appeared similar, whether the patients considered were aged < 5 years or ≥ 5 years, the development of a common case definition for use in both of these age groups appears reasonable.

Recently, two studies have evaluated case definitions for influenza in the developing countries of India and Kenya.27,28 The sensitivities (69–70%) and specificities (43–53%) recorded for the case definition for influenza-like illness in the present study are similar to the corresponding sensitivities (73–78%) and specificities (49–65%) recently reported in northern India, in a study that used the same broad enrolment criteria.27 However, the same case definition appeared less sensitive but more specific when investigated in Kenya, in a study confined to patients with acute respiratory illness.28

In the present study, 13% of the hospitalized patients aged < 5 years and 22% of those aged ≥ 5 years were found PCR-positive for influenza viruses. These relatively high prevalences may reflect the emergence of the novel pandemic A(H1N1)pdm09 influenza virus, the improved detection of patients with influenza (through the use of broad enrolment and testing criteria), and/or the effects of increased public and provider awareness of influenza, at least during the early pandemic period.30

The present study had several limitations. First, the findings apply only to hospitalized patients and not to outpatient or community settings. Second, despite the broad enrolment criteria, the proportion of enrolled patients with fever was high, indicating that infection is a leading cause of health-care seeking and hospitalization in our study population; this may not be the case in other settings. Third, an influenza pandemic occurred during the enrolment period. The investigated case definitions may have performed differently in the more common outbreaks of seasonal influenza, since the epidemiology (e.g. the age distribution of cases) and transmission dynamics of the seasonal and pandemic influenza viruses may well differ. Moreover, the panic created early on in the pandemic, especially in relation to the pandemic’s potential severity, could have altered the propensity of the general public to seek health care, at least during the first year of the present study. During influenza screening in several areas of China, the 2009 pandemic appears to have had a substantial impact on the apparent performance of case definitions of influenza-like illness.20,31 The current (post-pandemic) surveillance for influenza in Vadu DSA should allow the generalizability of the present results, to purely seasonal influenza, to be evaluated. Fourthly, it is likely that seasonal variation in influenza activity may influence the performance of case definitions. The present study covered a two-year period, however, and it is likely that such seasonality affects the predictive values of the case definitions more than their sensitivities or specificities. The positive and negative predictive values reported in this article were calculated only for the period when the prevalence of influenza virus infection among the investigated patients exceeded 10% (i.e. June 2009–October 2010). Lastly, the number of elderly patients (in whom influenza may manifest atypically, such as without the presence of fever) enrolled in the present study was very small.

The best choice of a case definition may vary according to the aim of the investigation. For diagnostic purposes, physicians may desire a clinical case definition with high predictive values, to guide the management of illness. For studies on vaccine efficacy, a case definition with higher specificity may be preferred. For case finding or rapid identification of influenza outbreaks, public-health professionals may prefer a case definition with high sensitivity and accept a definition with relatively low specificity. Our findings indicate that, in Vadu at least, the use of WHO’s standard case definition for severe acute respiratory illness, without adjusting for the definition’s low sensitivity, would result in a substantial underestimate of the burden posed by hospitalized cases of influenza. The use of a simplified version of this definition that does not include shortness of breath or, alternatively, the use of a parsimonious combination of symptoms and signs, such as “cough plus measured or reported fever” would provide a more accurate estimate. Our findings should help guide the choice of case definition, based on study or surveillance-system priorities.


Acknowledgements

The authors are grateful to Joshua Mott and Tim Uyeki for their input in the study design, to the private doctors and their organization “Spundun” and patients of the study hospitals, the field staff of Vadu DSA INDEPTH Network and the laboratory staff of the National Institute of Virology, for their participation in the study.

Funding:

The study reported forms part of a larger investigation (the Influenza Disease Burden, India, study), which was supported by the Centers for Disease Control and Prevention via a cooperative agreement (1U01IP000206).

Competing interests:

None declared.

References

Share