Terrill D. Hulson, MD

Influenza affects between 20% and 30% of the United States population annually. Three fourths of the infected individuals develop an acute respiratory illness, and one third of these seek medical attention. In a typical year, more than 20,000 to 40,000 Americans die, and more than 100,000 are hospitalized because of complications related to influenza.¹ Immunizations usually provide between 70% and 90% protection, yet many people at high risk do not receive the vaccine. An effective method for quickly differentiating patients with influenza from those with other respiratory illnesses might be helpful, since 2 new medications (the zanamivir inhaler and oseltamivir tablets) that are active against both influenza A and B are now available. Two other medications, amantadine and rimantidine, are active against influenza A only.^2,3 In the past, physicians have relied on symptoms alone or in conjunction with a manual or automated white blood cell (WBC) count with a differential count to assist them in making the diagnosis of influenza. However, several researchers have found that symptoms and signs have relatively poor predictive value during influenza outbreaks.^4-6 A recent pooled analysis of 3744 subjects involved in clinical trials of the antiviral agent zanamivir found that patients with influenza were somewhat more likely to have fever (68% vs 40%), cough (93% vs 80%), and nasal congestion (91% vs 81%) than patients with other infections.⁷

ZstatFlu is an office test for the diagnosis of both influenza A and B. It detects neuraminidase, an enzyme found on the influenza virus. In the presence of the virus, chromagen is cleaved off a synthetic neuraminidase substrate. A positive test results in a blue color change. The ZstatFlu test is 1 of 4 approved tests on the market for rapid diagnosis of influenza. Three of the 4 detect both influenza A and B⁸ To determine the most effective and efficient approach to the office diagnosis of influenza, we collected and analyzed clinical and laboratory data from patients seen in a family practice office setting during 2 consecutive influenza seasons.

Methods

Patients

We included consecutive patients presenting to a private family practice clinic that had 4 physicians and 1 physician assistant in a suburban setting in Oklahoma between January and March 1999 and November 1999 and January 2000 with fever, cough, sore throat, myalgia, and/or headache. No patients were systematically excluded. Consenting patients received a WBC and differential count, a rapid flu test (ZstatFlu), and an influenza culture by oropharynageal swab. Some patients consented to some but not all of these tests. Cultures were unavailable during a portion of each study period. Viral serologies were not done.

Procedure

Patients were triaged by the office nurse, who asked when they had become ill and whether they had experienced any of the following signs or symptoms: fever higher than 101°F (38.5 °C), cough, sore throat, headache, or myalgia. They were also asked if they had received that year’s flu vaccine. A WBC and differential count was performed by a laboratory technician in the clinic using a Cell-Dyn 1700 machine (Abbott Laboratories; Abbott Park, Ill). The ZstatFlu test was performed by the same laboratory technician using the method described by the manufacturer. An oropharyngeal swab for influenza culture was also obtained. The swabs were placed in viral culture medium and refrigerated. They were picked up once daily and transported to one of 2 laboratories (at ZymeTx or the Oklahoma State Department of Health) and plated that day. All participating patients gave informed consent. The protocol was approved by the Research Consultants Review Committee, Austin, Texas, and by the Institutional Review Committee at the University of Oklahoma Health Sciences Center.

Data Analysis

Only patients who had an influenza culture could be included in the analysis. Three separate influenza epidemics occurred during the 2 years of data collection. These outbreaks were first analyzed separately to evaluate consistency of results across epidemics and then as a combined data set for determination of overall test characteristics.

The following variables were considered: clinician, patient age, sex, duration of symptoms, delay in presentation, vaccine, cough, fever, myalgias, sore throat, headache, WBC count, differential WBC count, ZstatFlu result, and culture result. An additional variable, “flu symptoms,” was defined as the combination of fever, cough, and myalgia. Delay in presentation was further categorized as 2 days or less or more than 2 days, since treatment is most effective when begun within 2 days of the onset of illness. A left-shifted WBC count was defined arbitrarily as a polymorphonuclear leukocyte proportion greater than 60%, and a right-shifted WBC count was defined arbitrarily as a lymphocyte proportion greater than 40%.

Within each epidemic group, patients with positive cultures were compared with those who had negative cultures. Since the 2 influenza epidemics (A and B) during the first year occurred simultaneously, patients with negative cultures during that time were used for comparisons in both groups for these initial analyses. Comparisons were made for age and duration of symptoms using the Student’s t test for independent samples. All other comparisons were made using the chi-square statistic.

We combined all data. For these combined analyses, the influenza-negative patients from year 1 were counted only once. Receiver operating characteristic (ROC) curves were constructed for the ZstatFlu test, WBC count, and WBC count combined with the ZstatFlu test. Rockit 0.9B software (University of Chicago; Chicago, Ill) was used to determine the area under the curve (AUC) and confidence intervals for the WBC count and the WBC count combined with the ZstatFlu test by maximum likelihood estimation of the ROC parameters.⁹ Individual cut-points for WBC counts were compared as binary tests by calculating the AUC for each.¹⁰ To determine the AUC for the ZstatFlu test we used the nonparametric Wilcoxin statistic.¹¹ The logistic regression modeling function of Statistix7 software was used to analyze the individual and combined predictive properties of WBC count and ZstatFlu. Positive and negative likelihood ratios were calculated using standard formulas¹²; they correspond to the degree that a positive test result rules in disease and a negative test result rules out disease, respectively. These were used to estimate the rates of over- and undertreatment of influenza cases under 2 different baseline assumptions (pretest probabilities of influenza of 25% and 50%). Confidence intervals for sensitivity and specificity were calculated using the normal approximation to the binomial method.¹³

Results

We enrolled 382 patients during the first year (268 had influenza cultures performed) and 225 patients during the second year (90 were cultured). The total analyzable sample of cultured patients was 358 patients. In most cases, those who did not have cultures performed were seen on days when culture medium or laboratory pick-up were not available. Patients who had a culture performed were more likely to have a cough (P=.01) but otherwise did not differ from those who did not have a culture.In year 1, the influenza strains were A/Sydney (H3N2) and B/Bejing. In year 2, the strain was again A/Sydney (H3N2). The youngest patient with a positive flu culture was aged 10 months and the oldest was 73 years of age. The breakdown by age, sex, duration of symptoms, vaccine status, symptoms, WBC/differential, and ZstatFlu results by epidemic is shown in Table 1.

The presentation of influenza during the 3 epidemics differed. For example, the Beijing-like flu B in year 1 was more likely to infect younger people (mean age = 22.2 years) and was unlikely to cause a left WBC shift (25%), while the influenza A strain seen in the second year was more likely to infect older people (mean age = 28.3 years) and to be associated with a left WBC shift (72%). Culture-positive patients were somewhat more likely to report fever during 2 of the 3 outbreaks, but no single symptom or the symptom complex—fever, cough, and myalgias—reliably distinguished flu cases from nonflu cases across all epidemics.

Fifteen percent, 7%, and 17% of patients with positive influenza cultures in the 3 epidemics had received the vaccine. Both influenza strains were included in the vaccines given during those 2 years. However, immunization status was not consistently helpful for distinguishing influenza cases from those with other flu-like illnesses. Duration of symptoms was only associated with culture result in the year 2 flu A epidemic, in which influenza patients, on average, presented a half day earlier.

The WBC count was strongly associated with culture result in all 3 epidemics. As the WBC count increased, the likelihood of a positive culture decreased. A right or left shift in the differential count was not consistently related to the probability of a positive culture. WBC count was positively correlated with duration of symptoms in children (Pearson correlation coefficient = 0.20; P=.04) and negatively associated with symptom duration in adults (Pearson correlation coefficient = -0.15; P=.66). There was also a negative association between left shift and duration of symptoms (P=.001) and a positive association between right shift and duration of symptoms (P=.01) for all patients, suggesting that influenza patients develop a left shift at onset of infection and later convert to a right shift.

ROC curves were constructed using various levels of WBC counts with and without the ZstatFlu test Figure 1. For WBC count alone, the AUC was 0.67 (95% confidence interval [CI], 0.61-0.74). By comparison, the AUC for the ZstatFlu test was 0.74 (95% CI, 0.68-0.80). The ROC curve describing the use of a combination of ZstatFlu test and the WBC count had an AUC of 0.82 (95% CI, 0.76-0.87); this was better than WBC alone but not significantly different from ZstatFlu alone.

WBC counts greater than 7000 (negative likelihood ratio = 0.41) were superior to a negative ZstatFlu test result at confirming the absence of the flu. WBC counts less than 3200 (positive likelihood ratio = 7.21) were superior to a positive ZstatFlu test result at confirming the presence of the flu. A WBC count greater than 6300 had greater sensitivity (67%) than the ZstatFlu test, however, for WBC counts between 6300 and 7000, the gain in sensitivity did not offset the loss in specificity. A WBC count less than 4600 had a greater specificity (84%) than the ZstatFlu test, but for WBC counts between 3200 and 4600 the gain in specificity did not offset the loss in sensitivity.

Table 2 shows the characteristics of WBC counts at several cut-points, of the ZstatFlu test, and of their combinations. Using the one test strategy of treating those with a WBC count of 8000 or less would ensure treatment of almost all influenza cases (92%). Using the ZstatFlu test as a one testing strategy would assure that most of the patients treated have the flu but would miss 44% patients with the flu. Adding a WBC count if the ZstatFlu test result is negative improves sensitivity but reduces specificity. The predictive values positive and negative in the Table 1 are based on a previous probability of 50% (peak of flu season). These values would obviously be lower at the beginning or ending of an epidemic.

Discussion

Unfortunately, signs, symptoms, and vaccine status may be of little consistent value in distinguishing patients with influenza from those with other respiratory illnesses during influenza season. During some epidemics, fever and cough may be of some help, but this will depend heavily on what other illnesses are prevalent at the same time. Others have observed that symptoms have low predictive value and that physicians have difficulty identifying flu cases during epidemics.^5,6 Monto and colleagues⁷ reported that fever and cough occurred more frequently among influenza patients involved in clinical trials of an antiviral agent, but these results may not apply directly to primary care settings and represent pooled findings across several epidemics.

Vaccine status was also not helpful in this study for distinguishing influenza culture-positive patients. Influenza vaccination is effective in only 70% to 90% of patients. Therefore, there will always be vaccine-positive patients who develop the flu. Our data do not provide quantifiable information about overall vaccine efficacy, but the number of vaccine-positive patients was small, suggesting that the vaccine may have been effective in the community at large, though not in the culture-positive patients included in this study.

Both the WBC count and the ZstatFlu test can be helpful for identifying influenza cases. The testing strategy of choice depends to some degree on a number of factors including cost, duration and severity of symptoms, comorbidities, and potential adverse effects of treatment. The ZstatFlu test costs approximately $20. The cost of a WBC count is approximately $30, but it may have additional diagnostic value. Treating the patient with either zanamivir or oselfamivir costs $50 to $60, rimantidine $30, and amantadine $6.

The monetary value of an earlier return to work, reduced caregiver burden, or reduced transmission of infection will vary greatly. If the goal is to treat nearly every influenza case, a strategy of treating those with a WBC of 8000 or less appears to be the best strategy. If the goal is to be sure that only patients with the flu are treated, then treatment should be reserved for those who are ZstatFlu positive. Each patient and each physician would be expected to have different treatment thresholds that would affect the testing strategy. More than half the patients with positive influenza cultures were seen within 2 days of the onset of symptoms. These patients are the ones who would be most likely to benefit from the newer antiviral agents. For example, if the treatment threshold for a particular patient was 50%, no testing would have been necessary in any of the epidemics studied, since the pretest probabilities were all greater than 50%. An analysis that includes patient preferences would be helpful to determine the most cost-effective strategy.

The specificity of the ZstatFlu test is reported to be between 95% and 100%.⁴ However, when performed in this community family practice office by a laboratory technician trained by the test’s manufacturer, the specificity was only 85%. Many of the false-positive test results were coded as “weakly positive,” suggesting that the end point for positivity was somewhat unclear or that the laboratory technician was influenced by the patient’s symptoms. The specificity improved in the second year, suggesting an improvement in technique. We submit that this is an example of the discrepancy between test characteristics determined under “ideal” circumstances and test characteristics in actual practice settings. Another explanation is that patients with weakly positive ZstatFlu test results actually had influenza that would have been documented had serology been used as the gold standard instead of culture.

Limitations

A weakness of our study is the proportion of patients with flu-like symptoms from whom culture results were not available. Flu season actually began earlier than January during the 1998-1999 season and extended beyond January with the 1999-2000 season, but cultures were not available during a portion of these time periods. Although patients with cough were more likely to be cultured, this potential bias should not have affected our conclusions, since cough was not associated with culture result.

Two other diagnostic concerns are the lack of serologic tests and the known tendency of cultures to be more reliable early in the illness. Serology would probably have identified some additional influenza cases. This would have resulted in higher pretest probabilities of influenza. It is unclear how it would have affected the other analyses. Since there was no association between duration of symptoms and culture result in 2 of the 3 epidemics and in the combined analysis, we do not believe that waning sensitivity of flu cultures was a significant factor in this population. In the third epidemic it seems more likely that flu patients felt worse (eg, had more myalgias) and therefore came in earlier than that cultures became negative in those who delayed seeing the physician. Additionally, the study had insufficient power to detect a statistically significant difference between the diagnostic value of the ZstatFlu alone, and the combination of the WBC count and the ZstatFlu test.

It should be noted that patients were enrolled during an outbreak of influenza. In fact, the practice involved was one of the first in the state to recognize the onset of the epidemic because they were involved in this study. The conclusions reached about diagnostic strategies can only be generalized to similar epidemic situations.

Conclusions

Since influenza is associated with considerable morbidity and mortality, especially in high-risk populations, and given the brief window of opportunity (less than 48 hours) to treat patients with the flu with the newer agents, early and accurate diagnosis may be important in at least some cases. The use of screening WBC counts or rapid antigen tests could improve patient care during influenza epidemics. A cost-effectiveness analysis is needed to more fully elucidate this issue.

Acknowledgments

We would like to acknowledge the financial support of ZymeTx Corporation for supplying the ZstatFlu reagents, training, and influenza cultures at no cost. The support received from ZymeTx Corporation was unrestricted, and the company had no influence on our decision to analyze the results in the manner that we did or on the contents of this manuscript. We also want to thank Lavonne Glover for her expert assistance and patience in the preparation of the manuscript.

Influenza affects between 20% and 30% of the United States population annually. Three fourths of the infected individuals develop an acute respiratory illness, and one third of these seek medical attention. In a typical year, more than 20,000 to 40,000 Americans die, and more than 100,000 are hospitalized because of complications related to influenza.¹ Immunizations usually provide between 70% and 90% protection, yet many people at high risk do not receive the vaccine. An effective method for quickly differentiating patients with influenza from those with other respiratory illnesses might be helpful, since 2 new medications (the zanamivir inhaler and oseltamivir tablets) that are active against both influenza A and B are now available. Two other medications, amantadine and rimantidine, are active against influenza A only.^2,3 In the past, physicians have relied on symptoms alone or in conjunction with a manual or automated white blood cell (WBC) count with a differential count to assist them in making the diagnosis of influenza. However, several researchers have found that symptoms and signs have relatively poor predictive value during influenza outbreaks.^4-6 A recent pooled analysis of 3744 subjects involved in clinical trials of the antiviral agent zanamivir found that patients with influenza were somewhat more likely to have fever (68% vs 40%), cough (93% vs 80%), and nasal congestion (91% vs 81%) than patients with other infections.⁷

ZstatFlu is an office test for the diagnosis of both influenza A and B. It detects neuraminidase, an enzyme found on the influenza virus. In the presence of the virus, chromagen is cleaved off a synthetic neuraminidase substrate. A positive test results in a blue color change. The ZstatFlu test is 1 of 4 approved tests on the market for rapid diagnosis of influenza. Three of the 4 detect both influenza A and B⁸ To determine the most effective and efficient approach to the office diagnosis of influenza, we collected and analyzed clinical and laboratory data from patients seen in a family practice office setting during 2 consecutive influenza seasons.

Methods

Patients

We included consecutive patients presenting to a private family practice clinic that had 4 physicians and 1 physician assistant in a suburban setting in Oklahoma between January and March 1999 and November 1999 and January 2000 with fever, cough, sore throat, myalgia, and/or headache. No patients were systematically excluded. Consenting patients received a WBC and differential count, a rapid flu test (ZstatFlu), and an influenza culture by oropharynageal swab. Some patients consented to some but not all of these tests. Cultures were unavailable during a portion of each study period. Viral serologies were not done.

Procedure

Patients were triaged by the office nurse, who asked when they had become ill and whether they had experienced any of the following signs or symptoms: fever higher than 101°F (38.5 °C), cough, sore throat, headache, or myalgia. They were also asked if they had received that year’s flu vaccine. A WBC and differential count was performed by a laboratory technician in the clinic using a Cell-Dyn 1700 machine (Abbott Laboratories; Abbott Park, Ill). The ZstatFlu test was performed by the same laboratory technician using the method described by the manufacturer. An oropharyngeal swab for influenza culture was also obtained. The swabs were placed in viral culture medium and refrigerated. They were picked up once daily and transported to one of 2 laboratories (at ZymeTx or the Oklahoma State Department of Health) and plated that day. All participating patients gave informed consent. The protocol was approved by the Research Consultants Review Committee, Austin, Texas, and by the Institutional Review Committee at the University of Oklahoma Health Sciences Center.

Data Analysis

Only patients who had an influenza culture could be included in the analysis. Three separate influenza epidemics occurred during the 2 years of data collection. These outbreaks were first analyzed separately to evaluate consistency of results across epidemics and then as a combined data set for determination of overall test characteristics.

The following variables were considered: clinician, patient age, sex, duration of symptoms, delay in presentation, vaccine, cough, fever, myalgias, sore throat, headache, WBC count, differential WBC count, ZstatFlu result, and culture result. An additional variable, “flu symptoms,” was defined as the combination of fever, cough, and myalgia. Delay in presentation was further categorized as 2 days or less or more than 2 days, since treatment is most effective when begun within 2 days of the onset of illness. A left-shifted WBC count was defined arbitrarily as a polymorphonuclear leukocyte proportion greater than 60%, and a right-shifted WBC count was defined arbitrarily as a lymphocyte proportion greater than 40%.

Within each epidemic group, patients with positive cultures were compared with those who had negative cultures. Since the 2 influenza epidemics (A and B) during the first year occurred simultaneously, patients with negative cultures during that time were used for comparisons in both groups for these initial analyses. Comparisons were made for age and duration of symptoms using the Student’s t test for independent samples. All other comparisons were made using the chi-square statistic.

We combined all data. For these combined analyses, the influenza-negative patients from year 1 were counted only once. Receiver operating characteristic (ROC) curves were constructed for the ZstatFlu test, WBC count, and WBC count combined with the ZstatFlu test. Rockit 0.9B software (University of Chicago; Chicago, Ill) was used to determine the area under the curve (AUC) and confidence intervals for the WBC count and the WBC count combined with the ZstatFlu test by maximum likelihood estimation of the ROC parameters.⁹ Individual cut-points for WBC counts were compared as binary tests by calculating the AUC for each.¹⁰ To determine the AUC for the ZstatFlu test we used the nonparametric Wilcoxin statistic.¹¹ The logistic regression modeling function of Statistix7 software was used to analyze the individual and combined predictive properties of WBC count and ZstatFlu. Positive and negative likelihood ratios were calculated using standard formulas¹²; they correspond to the degree that a positive test result rules in disease and a negative test result rules out disease, respectively. These were used to estimate the rates of over- and undertreatment of influenza cases under 2 different baseline assumptions (pretest probabilities of influenza of 25% and 50%). Confidence intervals for sensitivity and specificity were calculated using the normal approximation to the binomial method.¹³

Results

We enrolled 382 patients during the first year (268 had influenza cultures performed) and 225 patients during the second year (90 were cultured). The total analyzable sample of cultured patients was 358 patients. In most cases, those who did not have cultures performed were seen on days when culture medium or laboratory pick-up were not available. Patients who had a culture performed were more likely to have a cough (P=.01) but otherwise did not differ from those who did not have a culture.In year 1, the influenza strains were A/Sydney (H3N2) and B/Bejing. In year 2, the strain was again A/Sydney (H3N2). The youngest patient with a positive flu culture was aged 10 months and the oldest was 73 years of age. The breakdown by age, sex, duration of symptoms, vaccine status, symptoms, WBC/differential, and ZstatFlu results by epidemic is shown in Table 1.

The presentation of influenza during the 3 epidemics differed. For example, the Beijing-like flu B in year 1 was more likely to infect younger people (mean age = 22.2 years) and was unlikely to cause a left WBC shift (25%), while the influenza A strain seen in the second year was more likely to infect older people (mean age = 28.3 years) and to be associated with a left WBC shift (72%). Culture-positive patients were somewhat more likely to report fever during 2 of the 3 outbreaks, but no single symptom or the symptom complex—fever, cough, and myalgias—reliably distinguished flu cases from nonflu cases across all epidemics.

Fifteen percent, 7%, and 17% of patients with positive influenza cultures in the 3 epidemics had received the vaccine. Both influenza strains were included in the vaccines given during those 2 years. However, immunization status was not consistently helpful for distinguishing influenza cases from those with other flu-like illnesses. Duration of symptoms was only associated with culture result in the year 2 flu A epidemic, in which influenza patients, on average, presented a half day earlier.

The WBC count was strongly associated with culture result in all 3 epidemics. As the WBC count increased, the likelihood of a positive culture decreased. A right or left shift in the differential count was not consistently related to the probability of a positive culture. WBC count was positively correlated with duration of symptoms in children (Pearson correlation coefficient = 0.20; P=.04) and negatively associated with symptom duration in adults (Pearson correlation coefficient = -0.15; P=.66). There was also a negative association between left shift and duration of symptoms (P=.001) and a positive association between right shift and duration of symptoms (P=.01) for all patients, suggesting that influenza patients develop a left shift at onset of infection and later convert to a right shift.

ROC curves were constructed using various levels of WBC counts with and without the ZstatFlu test Figure 1. For WBC count alone, the AUC was 0.67 (95% confidence interval [CI], 0.61-0.74). By comparison, the AUC for the ZstatFlu test was 0.74 (95% CI, 0.68-0.80). The ROC curve describing the use of a combination of ZstatFlu test and the WBC count had an AUC of 0.82 (95% CI, 0.76-0.87); this was better than WBC alone but not significantly different from ZstatFlu alone.

WBC counts greater than 7000 (negative likelihood ratio = 0.41) were superior to a negative ZstatFlu test result at confirming the absence of the flu. WBC counts less than 3200 (positive likelihood ratio = 7.21) were superior to a positive ZstatFlu test result at confirming the presence of the flu. A WBC count greater than 6300 had greater sensitivity (67%) than the ZstatFlu test, however, for WBC counts between 6300 and 7000, the gain in sensitivity did not offset the loss in specificity. A WBC count less than 4600 had a greater specificity (84%) than the ZstatFlu test, but for WBC counts between 3200 and 4600 the gain in specificity did not offset the loss in sensitivity.

Table 2 shows the characteristics of WBC counts at several cut-points, of the ZstatFlu test, and of their combinations. Using the one test strategy of treating those with a WBC count of 8000 or less would ensure treatment of almost all influenza cases (92%). Using the ZstatFlu test as a one testing strategy would assure that most of the patients treated have the flu but would miss 44% patients with the flu. Adding a WBC count if the ZstatFlu test result is negative improves sensitivity but reduces specificity. The predictive values positive and negative in the Table 1 are based on a previous probability of 50% (peak of flu season). These values would obviously be lower at the beginning or ending of an epidemic.

Discussion

Unfortunately, signs, symptoms, and vaccine status may be of little consistent value in distinguishing patients with influenza from those with other respiratory illnesses during influenza season. During some epidemics, fever and cough may be of some help, but this will depend heavily on what other illnesses are prevalent at the same time. Others have observed that symptoms have low predictive value and that physicians have difficulty identifying flu cases during epidemics.^5,6 Monto and colleagues⁷ reported that fever and cough occurred more frequently among influenza patients involved in clinical trials of an antiviral agent, but these results may not apply directly to primary care settings and represent pooled findings across several epidemics.

Vaccine status was also not helpful in this study for distinguishing influenza culture-positive patients. Influenza vaccination is effective in only 70% to 90% of patients. Therefore, there will always be vaccine-positive patients who develop the flu. Our data do not provide quantifiable information about overall vaccine efficacy, but the number of vaccine-positive patients was small, suggesting that the vaccine may have been effective in the community at large, though not in the culture-positive patients included in this study.

Both the WBC count and the ZstatFlu test can be helpful for identifying influenza cases. The testing strategy of choice depends to some degree on a number of factors including cost, duration and severity of symptoms, comorbidities, and potential adverse effects of treatment. The ZstatFlu test costs approximately $20. The cost of a WBC count is approximately $30, but it may have additional diagnostic value. Treating the patient with either zanamivir or oselfamivir costs $50 to $60, rimantidine $30, and amantadine $6.

The monetary value of an earlier return to work, reduced caregiver burden, or reduced transmission of infection will vary greatly. If the goal is to treat nearly every influenza case, a strategy of treating those with a WBC of 8000 or less appears to be the best strategy. If the goal is to be sure that only patients with the flu are treated, then treatment should be reserved for those who are ZstatFlu positive. Each patient and each physician would be expected to have different treatment thresholds that would affect the testing strategy. More than half the patients with positive influenza cultures were seen within 2 days of the onset of symptoms. These patients are the ones who would be most likely to benefit from the newer antiviral agents. For example, if the treatment threshold for a particular patient was 50%, no testing would have been necessary in any of the epidemics studied, since the pretest probabilities were all greater than 50%. An analysis that includes patient preferences would be helpful to determine the most cost-effective strategy.

The specificity of the ZstatFlu test is reported to be between 95% and 100%.⁴ However, when performed in this community family practice office by a laboratory technician trained by the test’s manufacturer, the specificity was only 85%. Many of the false-positive test results were coded as “weakly positive,” suggesting that the end point for positivity was somewhat unclear or that the laboratory technician was influenced by the patient’s symptoms. The specificity improved in the second year, suggesting an improvement in technique. We submit that this is an example of the discrepancy between test characteristics determined under “ideal” circumstances and test characteristics in actual practice settings. Another explanation is that patients with weakly positive ZstatFlu test results actually had influenza that would have been documented had serology been used as the gold standard instead of culture.

Limitations

A weakness of our study is the proportion of patients with flu-like symptoms from whom culture results were not available. Flu season actually began earlier than January during the 1998-1999 season and extended beyond January with the 1999-2000 season, but cultures were not available during a portion of these time periods. Although patients with cough were more likely to be cultured, this potential bias should not have affected our conclusions, since cough was not associated with culture result.

Two other diagnostic concerns are the lack of serologic tests and the known tendency of cultures to be more reliable early in the illness. Serology would probably have identified some additional influenza cases. This would have resulted in higher pretest probabilities of influenza. It is unclear how it would have affected the other analyses. Since there was no association between duration of symptoms and culture result in 2 of the 3 epidemics and in the combined analysis, we do not believe that waning sensitivity of flu cultures was a significant factor in this population. In the third epidemic it seems more likely that flu patients felt worse (eg, had more myalgias) and therefore came in earlier than that cultures became negative in those who delayed seeing the physician. Additionally, the study had insufficient power to detect a statistically significant difference between the diagnostic value of the ZstatFlu alone, and the combination of the WBC count and the ZstatFlu test.

It should be noted that patients were enrolled during an outbreak of influenza. In fact, the practice involved was one of the first in the state to recognize the onset of the epidemic because they were involved in this study. The conclusions reached about diagnostic strategies can only be generalized to similar epidemic situations.

Conclusions

Since influenza is associated with considerable morbidity and mortality, especially in high-risk populations, and given the brief window of opportunity (less than 48 hours) to treat patients with the flu with the newer agents, early and accurate diagnosis may be important in at least some cases. The use of screening WBC counts or rapid antigen tests could improve patient care during influenza epidemics. A cost-effectiveness analysis is needed to more fully elucidate this issue.

Acknowledgments

We would like to acknowledge the financial support of ZymeTx Corporation for supplying the ZstatFlu reagents, training, and influenza cultures at no cost. The support received from ZymeTx Corporation was unrestricted, and the company had no influence on our decision to analyze the results in the manner that we did or on the contents of this manuscript. We also want to thank Lavonne Glover for her expert assistance and patience in the preparation of the manuscript.

Influenza affects between 20% and 30% of the United States population annually. Three fourths of the infected individuals develop an acute respiratory illness, and one third of these seek medical attention. In a typical year, more than 20,000 to 40,000 Americans die, and more than 100,000 are hospitalized because of complications related to influenza.¹ Immunizations usually provide between 70% and 90% protection, yet many people at high risk do not receive the vaccine. An effective method for quickly differentiating patients with influenza from those with other respiratory illnesses might be helpful, since 2 new medications (the zanamivir inhaler and oseltamivir tablets) that are active against both influenza A and B are now available. Two other medications, amantadine and rimantidine, are active against influenza A only.^2,3 In the past, physicians have relied on symptoms alone or in conjunction with a manual or automated white blood cell (WBC) count with a differential count to assist them in making the diagnosis of influenza. However, several researchers have found that symptoms and signs have relatively poor predictive value during influenza outbreaks.^4-6 A recent pooled analysis of 3744 subjects involved in clinical trials of the antiviral agent zanamivir found that patients with influenza were somewhat more likely to have fever (68% vs 40%), cough (93% vs 80%), and nasal congestion (91% vs 81%) than patients with other infections.⁷

ZstatFlu is an office test for the diagnosis of both influenza A and B. It detects neuraminidase, an enzyme found on the influenza virus. In the presence of the virus, chromagen is cleaved off a synthetic neuraminidase substrate. A positive test results in a blue color change. The ZstatFlu test is 1 of 4 approved tests on the market for rapid diagnosis of influenza. Three of the 4 detect both influenza A and B⁸ To determine the most effective and efficient approach to the office diagnosis of influenza, we collected and analyzed clinical and laboratory data from patients seen in a family practice office setting during 2 consecutive influenza seasons.

Methods

Patients

We included consecutive patients presenting to a private family practice clinic that had 4 physicians and 1 physician assistant in a suburban setting in Oklahoma between January and March 1999 and November 1999 and January 2000 with fever, cough, sore throat, myalgia, and/or headache. No patients were systematically excluded. Consenting patients received a WBC and differential count, a rapid flu test (ZstatFlu), and an influenza culture by oropharynageal swab. Some patients consented to some but not all of these tests. Cultures were unavailable during a portion of each study period. Viral serologies were not done.

Procedure

Patients were triaged by the office nurse, who asked when they had become ill and whether they had experienced any of the following signs or symptoms: fever higher than 101°F (38.5 °C), cough, sore throat, headache, or myalgia. They were also asked if they had received that year’s flu vaccine. A WBC and differential count was performed by a laboratory technician in the clinic using a Cell-Dyn 1700 machine (Abbott Laboratories; Abbott Park, Ill). The ZstatFlu test was performed by the same laboratory technician using the method described by the manufacturer. An oropharyngeal swab for influenza culture was also obtained. The swabs were placed in viral culture medium and refrigerated. They were picked up once daily and transported to one of 2 laboratories (at ZymeTx or the Oklahoma State Department of Health) and plated that day. All participating patients gave informed consent. The protocol was approved by the Research Consultants Review Committee, Austin, Texas, and by the Institutional Review Committee at the University of Oklahoma Health Sciences Center.

Data Analysis

Only patients who had an influenza culture could be included in the analysis. Three separate influenza epidemics occurred during the 2 years of data collection. These outbreaks were first analyzed separately to evaluate consistency of results across epidemics and then as a combined data set for determination of overall test characteristics.

The following variables were considered: clinician, patient age, sex, duration of symptoms, delay in presentation, vaccine, cough, fever, myalgias, sore throat, headache, WBC count, differential WBC count, ZstatFlu result, and culture result. An additional variable, “flu symptoms,” was defined as the combination of fever, cough, and myalgia. Delay in presentation was further categorized as 2 days or less or more than 2 days, since treatment is most effective when begun within 2 days of the onset of illness. A left-shifted WBC count was defined arbitrarily as a polymorphonuclear leukocyte proportion greater than 60%, and a right-shifted WBC count was defined arbitrarily as a lymphocyte proportion greater than 40%.

Within each epidemic group, patients with positive cultures were compared with those who had negative cultures. Since the 2 influenza epidemics (A and B) during the first year occurred simultaneously, patients with negative cultures during that time were used for comparisons in both groups for these initial analyses. Comparisons were made for age and duration of symptoms using the Student’s t test for independent samples. All other comparisons were made using the chi-square statistic.

We combined all data. For these combined analyses, the influenza-negative patients from year 1 were counted only once. Receiver operating characteristic (ROC) curves were constructed for the ZstatFlu test, WBC count, and WBC count combined with the ZstatFlu test. Rockit 0.9B software (University of Chicago; Chicago, Ill) was used to determine the area under the curve (AUC) and confidence intervals for the WBC count and the WBC count combined with the ZstatFlu test by maximum likelihood estimation of the ROC parameters.⁹ Individual cut-points for WBC counts were compared as binary tests by calculating the AUC for each.¹⁰ To determine the AUC for the ZstatFlu test we used the nonparametric Wilcoxin statistic.¹¹ The logistic regression modeling function of Statistix7 software was used to analyze the individual and combined predictive properties of WBC count and ZstatFlu. Positive and negative likelihood ratios were calculated using standard formulas¹²; they correspond to the degree that a positive test result rules in disease and a negative test result rules out disease, respectively. These were used to estimate the rates of over- and undertreatment of influenza cases under 2 different baseline assumptions (pretest probabilities of influenza of 25% and 50%). Confidence intervals for sensitivity and specificity were calculated using the normal approximation to the binomial method.¹³

Results

We enrolled 382 patients during the first year (268 had influenza cultures performed) and 225 patients during the second year (90 were cultured). The total analyzable sample of cultured patients was 358 patients. In most cases, those who did not have cultures performed were seen on days when culture medium or laboratory pick-up were not available. Patients who had a culture performed were more likely to have a cough (P=.01) but otherwise did not differ from those who did not have a culture.In year 1, the influenza strains were A/Sydney (H3N2) and B/Bejing. In year 2, the strain was again A/Sydney (H3N2). The youngest patient with a positive flu culture was aged 10 months and the oldest was 73 years of age. The breakdown by age, sex, duration of symptoms, vaccine status, symptoms, WBC/differential, and ZstatFlu results by epidemic is shown in Table 1.

The presentation of influenza during the 3 epidemics differed. For example, the Beijing-like flu B in year 1 was more likely to infect younger people (mean age = 22.2 years) and was unlikely to cause a left WBC shift (25%), while the influenza A strain seen in the second year was more likely to infect older people (mean age = 28.3 years) and to be associated with a left WBC shift (72%). Culture-positive patients were somewhat more likely to report fever during 2 of the 3 outbreaks, but no single symptom or the symptom complex—fever, cough, and myalgias—reliably distinguished flu cases from nonflu cases across all epidemics.

Fifteen percent, 7%, and 17% of patients with positive influenza cultures in the 3 epidemics had received the vaccine. Both influenza strains were included in the vaccines given during those 2 years. However, immunization status was not consistently helpful for distinguishing influenza cases from those with other flu-like illnesses. Duration of symptoms was only associated with culture result in the year 2 flu A epidemic, in which influenza patients, on average, presented a half day earlier.

The WBC count was strongly associated with culture result in all 3 epidemics. As the WBC count increased, the likelihood of a positive culture decreased. A right or left shift in the differential count was not consistently related to the probability of a positive culture. WBC count was positively correlated with duration of symptoms in children (Pearson correlation coefficient = 0.20; P=.04) and negatively associated with symptom duration in adults (Pearson correlation coefficient = -0.15; P=.66). There was also a negative association between left shift and duration of symptoms (P=.001) and a positive association between right shift and duration of symptoms (P=.01) for all patients, suggesting that influenza patients develop a left shift at onset of infection and later convert to a right shift.

ROC curves were constructed using various levels of WBC counts with and without the ZstatFlu test Figure 1. For WBC count alone, the AUC was 0.67 (95% confidence interval [CI], 0.61-0.74). By comparison, the AUC for the ZstatFlu test was 0.74 (95% CI, 0.68-0.80). The ROC curve describing the use of a combination of ZstatFlu test and the WBC count had an AUC of 0.82 (95% CI, 0.76-0.87); this was better than WBC alone but not significantly different from ZstatFlu alone.

WBC counts greater than 7000 (negative likelihood ratio = 0.41) were superior to a negative ZstatFlu test result at confirming the absence of the flu. WBC counts less than 3200 (positive likelihood ratio = 7.21) were superior to a positive ZstatFlu test result at confirming the presence of the flu. A WBC count greater than 6300 had greater sensitivity (67%) than the ZstatFlu test, however, for WBC counts between 6300 and 7000, the gain in sensitivity did not offset the loss in specificity. A WBC count less than 4600 had a greater specificity (84%) than the ZstatFlu test, but for WBC counts between 3200 and 4600 the gain in specificity did not offset the loss in sensitivity.

Table 2 shows the characteristics of WBC counts at several cut-points, of the ZstatFlu test, and of their combinations. Using the one test strategy of treating those with a WBC count of 8000 or less would ensure treatment of almost all influenza cases (92%). Using the ZstatFlu test as a one testing strategy would assure that most of the patients treated have the flu but would miss 44% patients with the flu. Adding a WBC count if the ZstatFlu test result is negative improves sensitivity but reduces specificity. The predictive values positive and negative in the Table 1 are based on a previous probability of 50% (peak of flu season). These values would obviously be lower at the beginning or ending of an epidemic.

Discussion

Unfortunately, signs, symptoms, and vaccine status may be of little consistent value in distinguishing patients with influenza from those with other respiratory illnesses during influenza season. During some epidemics, fever and cough may be of some help, but this will depend heavily on what other illnesses are prevalent at the same time. Others have observed that symptoms have low predictive value and that physicians have difficulty identifying flu cases during epidemics.^5,6 Monto and colleagues⁷ reported that fever and cough occurred more frequently among influenza patients involved in clinical trials of an antiviral agent, but these results may not apply directly to primary care settings and represent pooled findings across several epidemics.

Vaccine status was also not helpful in this study for distinguishing influenza culture-positive patients. Influenza vaccination is effective in only 70% to 90% of patients. Therefore, there will always be vaccine-positive patients who develop the flu. Our data do not provide quantifiable information about overall vaccine efficacy, but the number of vaccine-positive patients was small, suggesting that the vaccine may have been effective in the community at large, though not in the culture-positive patients included in this study.

Both the WBC count and the ZstatFlu test can be helpful for identifying influenza cases. The testing strategy of choice depends to some degree on a number of factors including cost, duration and severity of symptoms, comorbidities, and potential adverse effects of treatment. The ZstatFlu test costs approximately $20. The cost of a WBC count is approximately $30, but it may have additional diagnostic value. Treating the patient with either zanamivir or oselfamivir costs $50 to $60, rimantidine $30, and amantadine $6.

The monetary value of an earlier return to work, reduced caregiver burden, or reduced transmission of infection will vary greatly. If the goal is to treat nearly every influenza case, a strategy of treating those with a WBC of 8000 or less appears to be the best strategy. If the goal is to be sure that only patients with the flu are treated, then treatment should be reserved for those who are ZstatFlu positive. Each patient and each physician would be expected to have different treatment thresholds that would affect the testing strategy. More than half the patients with positive influenza cultures were seen within 2 days of the onset of symptoms. These patients are the ones who would be most likely to benefit from the newer antiviral agents. For example, if the treatment threshold for a particular patient was 50%, no testing would have been necessary in any of the epidemics studied, since the pretest probabilities were all greater than 50%. An analysis that includes patient preferences would be helpful to determine the most cost-effective strategy.

The specificity of the ZstatFlu test is reported to be between 95% and 100%.⁴ However, when performed in this community family practice office by a laboratory technician trained by the test’s manufacturer, the specificity was only 85%. Many of the false-positive test results were coded as “weakly positive,” suggesting that the end point for positivity was somewhat unclear or that the laboratory technician was influenced by the patient’s symptoms. The specificity improved in the second year, suggesting an improvement in technique. We submit that this is an example of the discrepancy between test characteristics determined under “ideal” circumstances and test characteristics in actual practice settings. Another explanation is that patients with weakly positive ZstatFlu test results actually had influenza that would have been documented had serology been used as the gold standard instead of culture.

Limitations

A weakness of our study is the proportion of patients with flu-like symptoms from whom culture results were not available. Flu season actually began earlier than January during the 1998-1999 season and extended beyond January with the 1999-2000 season, but cultures were not available during a portion of these time periods. Although patients with cough were more likely to be cultured, this potential bias should not have affected our conclusions, since cough was not associated with culture result.

Two other diagnostic concerns are the lack of serologic tests and the known tendency of cultures to be more reliable early in the illness. Serology would probably have identified some additional influenza cases. This would have resulted in higher pretest probabilities of influenza. It is unclear how it would have affected the other analyses. Since there was no association between duration of symptoms and culture result in 2 of the 3 epidemics and in the combined analysis, we do not believe that waning sensitivity of flu cultures was a significant factor in this population. In the third epidemic it seems more likely that flu patients felt worse (eg, had more myalgias) and therefore came in earlier than that cultures became negative in those who delayed seeing the physician. Additionally, the study had insufficient power to detect a statistically significant difference between the diagnostic value of the ZstatFlu alone, and the combination of the WBC count and the ZstatFlu test.

It should be noted that patients were enrolled during an outbreak of influenza. In fact, the practice involved was one of the first in the state to recognize the onset of the epidemic because they were involved in this study. The conclusions reached about diagnostic strategies can only be generalized to similar epidemic situations.

Conclusions

Since influenza is associated with considerable morbidity and mortality, especially in high-risk populations, and given the brief window of opportunity (less than 48 hours) to treat patients with the flu with the newer agents, early and accurate diagnosis may be important in at least some cases. The use of screening WBC counts or rapid antigen tests could improve patient care during influenza epidemics. A cost-effectiveness analysis is needed to more fully elucidate this issue.

Acknowledgments

We would like to acknowledge the financial support of ZymeTx Corporation for supplying the ZstatFlu reagents, training, and influenza cultures at no cost. The support received from ZymeTx Corporation was unrestricted, and the company had no influence on our decision to analyze the results in the manner that we did or on the contents of this manuscript. We also want to thank Lavonne Glover for her expert assistance and patience in the preparation of the manuscript.