Influenza Burden of Illness: Estimates From a National Prospective Survey of Household Contacts in France

Background  The burden of influenza among ambulatory patients is still relatively unknown, although this knowledge is crucial for evaluating strategies against influenza. We estimated the impact of influenza in terms of uncomplicated morbidity and its consequences on health care utilization and lost workdays.

Methods  A national prospective household contact study between January 4, 2000, and March 15, 2000, in France recruited the households of 946 persons who visited a physician (index cases); 395 households with influenza-positive index cases completed the follow-up, which assessed the clinical impact of influenza, medical visits, treatment, and lost workdays in these index cases and their contacts.

Results  Of 817 assessable household contacts, 313 developed clinical influenza (secondary cases); 178 (57%) of them visited a physician at least once (consulting secondary cases). The median duration of illness was 8 days (95% confidence interval [CI], 7-8 days) in index cases, 7 days (95% CI, 7-8 days) in consulting secondary cases, and 4 days (95% CI, 3-5 days) in nonconsulting secondary cases (P<.001); the median duration of treatment in these groups was 8 days (95% CI, 8-9 days), 8 days (95% CI, 7-10 days), and 5 days (95% CI, 4-6 days), respectively (P<.001); and their mean ± SD number of lost workdays was 4.0 ± 2.8, 2.9 ± 2.5, and 0.3 ± 0.6, respectively, in working adults (P<.001).

Conclusions  These results confirm the substantial burden of illness of influenza. The results should be useful for evaluating the cost-effectiveness of strategies against influenza.

INFLUENZA EPIDEMICS have an important and well-documented impact on mortality, particularly in at-risk individuals, including elderly persons.^1-4 Recent work⁵ suggests that influenza epidemics are also associated with an elevated number of hospitalizations. On the other hand, their impact on uncomplicated morbidity and their consequences on health care utilization and lost productivity are still relatively unclear, despite the development of influenza surveillance systems in many countries. These systems are based on data collected from physicians who report cases of syndromes meeting a clinical case definition of influenza, sometimes but not always combined with virologic testing; they make it possible to detect epidemics early, to identify the circulating viral strains, and to supply quantitative information, for example, about the number of "excess" physician visits associated with an epidemic.^6,7 For several reasons, however, the results of this surveillance yield an incomplete picture of the true impact of epidemics. The clinical expression of the infection caused by influenza viruses is variable, and some physician visits for influenza infection will not be counted (and no sample will be taken) because the patient's clinical presentation does not meet the case definition. Carrat et al⁸ previously showed that nearly half of the patients visiting a physician for influenza had an attenuated form of the disease. Moreover, some individuals with influenza do not visit physicians; this fraction may represent 30% to 65% of patients whose clinical picture suggests influenza.^9,10 These individuals cannot be captured by a medically based surveillance system. At the other end of the spectrum, some patients may visit a physician several times for the same influenza episode,¹¹ and these new visits are rarely counted. Failure of the surveillance systems to take into account these categories of patients leads to a potentially important underestimation of the epidemic's impact on uncomplicated morbidity. This knowledge is nonetheless essential, especially for an accurate evalua tion of the clinical and economic benefits of preventive and treatment strategies against influenza.^12,13

A prospective household contact study was performed between January 4, 2000, and March 15, 2000, to describe the different syndromes associated with influenza virus entry into a household and their consequences on health care utilization and lost workdays among patients who did and did not visit physicians.

The study took place during an influenza epidemic associated with the circulation of 2 viruses: A/Sydney/5/97 (H3N2) (the variant responsible for epidemics in Europe since the winter of 1997-1998) and A/Moscow/10/99 (H3N2) (a variant antigenically closely related to the previous one). Households were included in this study by 161 general practitioners across France who were members of the "Sentinelles" network.¹⁴ These physicians received training in taking virologic samples and about this study during a pilot phase in October 1999.

A household was recruited when 1 member (the index case) made an office visit to or received a house call from a physician, and a household was included in the study if the following criteria were met: the index case visited a participating physician because of the onset within 48 hours of a fever (temperature >38°C) or feverishness, with respiratory signs, and the index case lived with at least 1 other person (the contact subject), was the first case in the household, was not hospitalized as a result of this first visit, and consented to virologic sampling and study participation.

During the visit, the physician explained the principle of the study to an adult household member (the index case or the adult accompanying a minor index case), took a nasal swab sample from the index case and mailed it to 1 of the 3 participating virology centers, and mailed an inclusion form to the study administrative center. The physician also gave the adult a 2-part questionnaire to be completed at home for all members of the household: the first part collected the general characteristics, the influenza and vaccination history, and the social and demographic characteristics of each household member, and the second part, presented as a diary checklist, described for each person on each of the following 15 days the following items: symptoms (13 symptoms and "none of the above"), work, medications, and physician visits. If any family member visited a physician, information about any prescriptions and tests ordered were noted on a separate sheet. The questionnaire was to be mailed to the study center after 15 days of follow-up.

To include the broadest possible spectrum of clinical syndromes occurring among the contact subjects, we defined clinical influenza by the presence within 5 days of inclusion of fever (temperature >38°C) or feverishness (patient's subjective sensation of fever or chills) or at least 2 of the following signs: sore throat, headache, stiffness or myalgias, fatigue, cough, and nasal congestion, rhinorrhea, or sneezing. To avoid modifying the contact subjects' behavior as to health care utilization, no samples were obtained from them.

In the 3 days after inclusion, the study administration center telephoned 92% of the households to offer help in completing the questionnaire. The first 528 households were also enrolled in a study that measured their willingness to pay for faster amelioration of the principal influenza symptoms.

Nasal swab samples were collected and stored in 3 mL of viral transportation medium. Samples were sent through the mail without any refrigeration. Mean ± SD transportation time was 1.7 ± 1.4 days. One milliliter of sample was used immediately for direct immunofluorescence staining. The remaining 2 mL was stored at −70°C until use for culture and polymerase chain reaction (PCR).

One milliliter of sample was used to perform direct immunofluorescence staining on the cell pellets with specific monoclonal antibodies for influenza A and B viruses, parainfluenzae viruses, adenoviruses, and respiratory syncytial viruses (DAKO Imagen, Trappes, France). Viral cultures were performed on Madin-Darby canine kidney cell monolayers.

Two types of PCR were used to test for influenza viruses. Reverse transcription PCR with hybridization was used by 2 of the 3 virology centers; it has been validated and described elsewhere.¹⁵ In the third virology center, a real-time PCR (TaqMan; Applied Biosystems, Foster City, Calif) followed the method recently reported¹⁶ but used other specific primers. A cross-validation study of these different PCR tests is under way. Influenza was diagnosed if the results of any 1 of these 3 tests were positive.

The impact of influenza was estimated by identifying 3 groups: index cases, secondary cases who visited a physician at least once (consulting), and secondary cases who did not visit a physician (nonconsulting). Each participant checked off his or her symptoms daily from the following list: fever (temperature >38°C); feverishness; cough; sore throat; nasal congestion, rhinorrhea, or sneezing; dysphonia; fatigue; headache; stiffness or myalgias; otalgias; ocular symptoms; loss of appetite; and sleep disturbances. A symptom score was calculated by dividing the total number of symptoms by 13: a score of 1 indicated the presence of all 13 symptoms, and a score of 0 indicated the absence of any symptoms. To estimate the time lost from work by working participants, a daily index was calculated: a normal workday was coded 0, a complete workday in which work was disturbed because of illness was coded 0.25, absence of less than 1 full workday was coded 0.5, and a complete day of absence was coded 1. Weekends and holidays were coded 0. This index gives us an idea of the real time lost from work associated with influenza-related sick leave.

Quantitative variables were compared between groups by using analysis of variance. Qualitative variables were compared using the χ² test or, for small groups, an exact test. Correlation coefficients were estimated using the Spearman method. Median durations were estimated using the Kaplan-Meier method, and comparisons of duration were made using a log-rank test. Repeated-measures analyses of variance were used to compare quantitative variables measured over time, and the time-variable interactions were tested. All comparisons used bilateral tests and a 5% cutoff point. Data are given as mean ± SD.

Between January 4, 2000, and March 15, 2000, 946 households throughout France were included in the study; in 510 households (54%), the index case had positive test results for influenza A (Figure 1). Two index cases were infected by influenza B (1 of whom was also positive for type A) and 25 by respiratory syncytial virus (2 of whom were also positive for influenza A). There were no positive diagnoses of adenovirus or of any parainfluenza virus. One sample, which arrived at the laboratory 29 days after mailing, was not analyzed. The virologic examination results of 411 index cases (43%) were negative for all of the viruses tested.

Of the 510 households with an index case diagnosed as having influenza A, 395 (77%) returned the questionnaire and 115 (23%) were lost to follow-up. The percentage of households lost to follow-up varied according to the sex of the index case (19% for women and 26% for men; P = .04). At inclusion, households that completed the study and those lost to follow-up did not differ significantly for the other variables, in particular, for age of the index case, severity of the disease as reported by the physician, temperature measured by the physician or reported by the patient, inclusion date, drugs prescribed during the visit, or proportion of working adults among the index cases.

The 395 households we followed had a mean size (including the index case) of 3.25 ± 1.26 individuals. Insufficient information about the clinical follow-up meant that 74 of 891 contact subjects could not be categorized. Of the others, 313 (38%) developed symptoms in the 5 days after inclusion and were classified as secondary cases. At least 1 secondary case was reported in 56% of the households in which the index case was positive for influenza A.

The mean age of the individuals followed was 32.8 ± 19.7 years (range, 0.2-89.9 years), 49% were men, 10% had had an influenzalike illness during the preceding winter, 8% had been vaccinated against influenza, 19% were current smokers, 46% were working adults, and 16% had 1 or more chronic diseases (36% cardiovascular, 27% pulmonary, 3% diabetes mellitus, and 40% others). Table 1 gives the principal characteristics of the individuals we followed as a function of their status. Index cases were, on average, older and more frequently affected by chronic diseases than contacts. Contacts who became secondary cases did not differ significantly from those who remained healthy.

Of the 313 secondary cases, 178 (57%) visited a physician at least once for their symptoms (consulting) and 135 (43%) did not (nonconsulting).

Table 2 describes the clinical presentation of the syndromes observed, according to group. The mean time elapsed between inclusion and appearance of the first symptoms was 1.35 ± 1.50 days, with no difference between consulting and nonconsulting cases. On the first day of illness, the signs most frequently found in secondary cases were coughing and fatigue, but only 44% of the consulting and 30% of the nonconsulting secondary cases reported a fever or feverishness that day (P = .02). The median duration until alleviation of major symptoms in index cases was 8 days (95% confidence interval [CI], 7-8 days), which did not differ significantly from that in consulting secondary cases (7 days; 95% CI, 7-8 days) but was greater than in nonconsulting secondary cases (4 days; 95% CI, 3-5 days; P<.001) (Figure 2).

Table 3 describes the utilization of health care by group. The mean number of visits related to clinical influenza was 1.35 ± 0.63 in index cases and 1.35 ± 0.66 in consulting secondary cases (P = .91). Secondary cases first visited a physician a mean of 1.1 ± 2.0 days after symptoms began. Second visits, if any, took place 4.8 ± 3.2 days after the first. The mean number of prescription and over-the-counter drugs taken on the first day of illness differed among groups; this number was correlated with the clinical severity of the patient's disease (r = 0.41; P<.001). The proportion of patients treated with antibiotics varied with the number of visits: 17% for nonconsulting secondary cases, 48% for those who visited a physician only once, 70% for those consulting twice, and 85% for those who visited a physician more than twice (P<.001, Cochran-Armitage trend χ²). The median duration of antibiotic treatment was 6 days (95% CI, 6-7 days) in index cases, 7 days (95% CI, 6-8 days) in consulting secondary cases, and 4.5 days (95% CI, 3-6 days) in nonconsulting secondary cases (P = .02). In all, the median duration of actual treatment, whatever the drug, was higher in consulting (8 days) than in nonconsulting (5 days) secondary cases (P<.001) and was correlated with the symptom score on the first day of the disease (hazard ratio, 0.54 [95% CI, 0.38-0.77]; P<.001). The percentage of medicine prescribed for the patient among all the medicine taken was 90% for index cases, 85% for consulting secondary cases, and 28% for nonconsulting secondary cases (P<.001). In the latter, this result is explained by the use of treatments prescribed for them during the visit of another member of the household or of medication prescribed for them before study entry. The following additional tests were ordered: 15 lung radiographs, 5 sinus radiographs, and 1 complete blood cell count.

Working adults accounted for 199 of the index cases (50%), 66 of the consulting secondary cases (37%), and 65 of the nonconsulting secondary cases (48%). Figure 3 shows the daily index of sick leave from work for the 15 days after the disease began. The 3 groups differ in the mean number of lost workdays (sum of the daily index for the 15 days for each group), the percentage of individuals who did not stop working, and the median duration until return to usual work activity in those who had stopped working (Table 4). The difference is greatest between individuals who visited physicians and those who did not. The number of lost workdays was correlated with the symptom score on the first day of illness (r = 0.45; P<.001). Of working contacts who did not become secondary cases, the number of lost workdays was estimated to be 0.03 ± 0.18 days per sick adult in the household and 0.11 ± 0.41 days per sick child.

The results of this study confirm that the impact of the uncomplicated morbidity associated with influenza epidemics is substantial and cannot be deduced directly from surveillance data. We showed that 43% of the contact subjects who experienced an episode of respiratory tract infection, presumably caused by influenza, did not visit a physician for these episodes; that the mean number of visits for those who did go to a physician was 1.35; and that the duration of treatment, like the number of drugs taken and the translation of the episodes into lost workdays, varied as a function of recourse to consultation, which in turn was associated with the seriousness of the patient's illness.

Of the principal choices and assumptions that might have affected the validity of our results, we must discuss the potential bias induced by the choices of our experimental design, especially in terms of the representativeness of the population studied and of the case definition retained for classifying contact subjects.

We used a quantitative approach to assess uncomplicated morbidity by performing a prospective study of household contacts. The advantage of this type of design over a standard cohort design is that it increases the study power at the same time as it reduces the duration of follow-up for each case and increases the likelihood that the study cases will develop the disease. On the other hand, it has the disadvantage of preventing observation of individuals who live alone. Nonetheless, we do not believe that exclusion of these individuals biased our results substantially. The principal factor correlated in this study with utilization of health care resources and lost workdays was the clinical seriousness of the patient's illness, and there is no reason to think that this might be correlated with living alone: no correlation was observed between household size and the secondary cases' symptom score on the first day of illness (r = −0.04; P = .44).

We compared participants with the last census for several variables to assess national representativeness. The mean size of the households we studied was 3.25 persons, and the mean age of household members was 33 years; the corresponding national figures for households with more than 1 person are similar (3.10 persons and 34 years). The proportion of working adults (among all study participants) was 46%, whereas the national proportion is 45%. The social and demographic structure of the households in this study do not differ in any important way from the national means for occupational category or mean net household income. Moreover, all French regions were represented. Thus, it seems reasonable to consider that the households observed were representative of the French population.

The other possible source of bias is that the participants' inclusion in the study might have induced a change in their behavior as to health care utilization (drugs and physician visits). This bias is inherent in all prospective studies, but its extent is less herein than that observed in therapeutic trials. We intentionally provided no instructions to either the physician or the family about recommended treatments, physician visits, and so on. For the same reason, virologic samples were not taken from contact subjects who developed symptoms, and follow-up information was collected by a self-administered questionnaire. Our results for the prescription of antibiotic treatment are consistent with the values reported in France (61% of antibiotic use, data from a national retrospective survey of a representative population sample in 1999/2000¹⁷). The rate of neuraminidase inhibitor prescriptions was also consistent with data for their sales in France during the 1999/2000 epidemic (A. El-Hasnaoui, MD, PhD, Health Economic Department, GlaxoSmithKline, oral communication, December 2000). For these reasons, we do not think that the results were biased by the participants' participation in the follow-up. Households enrolled in the willingness-to-pay nested study did not differ from the others in terms of rate of secondary cases, self-evaluation of symptoms, and rate of physician visits; these data indicate that the results were not affected by participation in the willingness-to-pay study.

We deliberately used a "broad" definition of secondary cases that allowed inclusion of patients with minor clinical forms so that we could accurately quantify the impact of influenza, and we did not collect virologic samples from contacts for the reasons stated previously. It is accordingly possible that some contacts classified as secondary cases were infected by a respiratory pathogen other than an influenza virus. It also explains why we observed a substantially higher proportion of secondary cases among contact subjects of positive index cases (38%) than the corresponding proportions of laboratory-confirmed clinical influenza recently reported in the placebo groups of randomized studies of prophylaxis for contact subjects (13%-15%).^18,19 In these studies, the case definitions used for clinical influenza relied on a combination of symptoms sufficiently severe that 35% and 39% of laboratory-confirmed influenza infections among contacts did not have clinical influenza. In one of these studies,¹⁸ low-grade illnesses that did not meet the case definition represented 28% of all symptomatic influenza infections. Applying the case definitions used in those studies to our patients showed that they would not have classified 74 of our secondary cases (24%) as clinical influenza, mainly because of lack of fever. In other words, we found a similar rate of low-grade illnesses in our study. Moreover, our 38% rate of secondary cases is lower than that observed in the control group of another trial²⁰ in which the onset of a respiratory tract illness in contacts of nonvaccinated children was 56%; it fell to 20% when fever was included in the disease definition criteria. A temperature higher than 37.8°C is a relatively erratic sign among patients with positive test results for influenza: 30% of the positive individuals in the zanamivir trial²¹ did not meet this inclusion criterion. Two other studies^9,22 found that 35% and 40% of influenza A/H3N2–positive patients were afebrile. Thus, we believe that the spectrum of syndromes classified as secondary cases in our study corresponded to the spectrum of syndromes associated with influenza infections.

Although contact cases did not undergo virologic testing, we think that we minimized the risk that the syndromes observed in secondary cases had a noninfluenza origin by choosing a window of 5 days from the inclusion of the index case positive for influenza A for the onset of symptoms among secondary cases. This conclusion is also supported by the actual mean delay to onset of symptoms of 1.3 days. Finally, the combination of symptoms that we used to determine secondary cases was different from the inclusion criteria that justified virologic samples from the index cases; this difference is seen in the higher symptom scores in index cases vs secondary cases.

To our knowledge, there are no recent observational studies in general populations that would allow us to compare these results with other data. The median duration of disease observed in this study in individuals who visited a physician was similar to that reported for placebo groups in neuraminidase inhibitor trials,²³ although lower values have been observed.^24,25 The rate of physician visits by individuals with documented influenza varied between 38% and 49% in observational studies conducted in the United States between 1976 and 1980.²⁶ Our results are close to those observed in these studies.

The impact on productivity, that is, workdays lost or wasted because of influenza, is the key indicator in an assessment of the economic benefits of influenza control strategies—either treatment or vaccination. The mean number of days lost per working adult has been assessed to be 2.8 to 10.0 principally on the basis of studies in occupational settings^11,27 and 2.0 to 5.0 in pandemic contexts,²⁸ although lower values have also been observed in the placebo groups in prevention studies.^29-31 Our results are, on the whole, consistent with these values.

In conclusion, this study is, to our knowledge, the first to describe the repercussions on health care utilization and work that follow influenza virus entry into households. It shows that a substantial fraction of the population of patients with influenza "escapes" standard surveillance. We believe that these results provide important data for evaluating the cost-effectiveness of strategies against influenza.

Accepted for publication January 3, 2002.

This study was funded by GlaxoSmithKline, Marly-le Roi, France.

Preliminary analyses of these data were presented as abstract W21-4 at the Congress Options for the Control of Influenza IV, Hersonissos, Greece, September 24, 2000.

We thank the patients, their families, and the 161 general practitioners of the French "Sentinelles" system who participated in the study; Isabelle Goderel, BS, Gregory Pannetier, BS, Audrey Lavenu, MSc, Michael Schwarzinger, MD, MSc, and Astrid Vabret, MD, for technical assistance in data collection and analysis; Bertrand Basset, BS, Hélène Chabernaud, BS, and Anna Garafano, BS, for technical assistance in data processing; and Abdelkader El-Hasnaoui, MD, PhD (GlaxoSmithKline) for helpful comments.