Conceived and designed the experiments: KHM AF DE BK LT. Analyzed the data: KJK KHM. Wrote the paper: JH KHM. Planned data analysis strategy: JH KJK KHM AF DE BK LT. Contributed to writing of paper: KJK AF DE LT. Designed and conducted surveillance for deaths and hospitalizations due to nH1N1: AF.
JLH is a member of the Center for Disease Control and Prevention's Coordinating Center for Infectious Diseases Board of Scientific Counselors. However, this work was done independently of that position and the author does do not believe there is a true conflict.
The public health response to pandemic influenza is contingent on the pandemic strain's severity. In late April 2009, a potentially pandemic novel H1N1 influenza strain (nH1N1) was recognized. New York City (NYC) experienced an intensive initial outbreak that peaked in late May, providing the need and opportunity to rapidly quantify the severity of nH1N1.
Telephone surveys using rapid polling methods of approximately 1,000 households each were conducted May 20–27 and June 15–19, 2009. Respondents were asked about the occurrence of influenza-like illness (ILI, fever with either cough or sore throat) for each household member from May 1–27 (survey 1) or the preceding 30 days (survey 2). For the overlap period, prevalence data were combined by weighting the survey-specific contribution based on a Serfling model using data from the NYC syndromic surveillance system. Total and age-specific prevalence of ILI attributed to nH1N1 were estimated using two approaches to adjust for background ILI: discounting by ILI prevalence in less affected NYC boroughs and by ILI measured in syndromic surveillance data from 2004–2008. Deaths, hospitalizations and intensive care unit (ICU) admissions were determined from enhanced surveillance including nH1N1-specific testing. Combined ILI prevalence for the 50-day period was 15.8% (95% CI:13.2%–19.0%). The two methods of adjustment yielded point estimates of nH1N1-associated ILI of 7.8% and 12.2%. Overall case-fatality (CFR) estimates ranged from 0.054–0.086 per 1000 persons with nH1N1-associated ILI and were highest for persons ≥65 years (0.094–0.147 per 1000) and lowest for those 0–17 (0.008–0.012). Hospitalization rates ranged from 0.84–1.34 and ICU admission rates from 0.21–0.34 per 1000, with little variation in either by age-group.
ILI prevalence can be quickly estimated using rapid telephone surveys, using syndromic surveillance data to determine expected “background” ILI proportion. Risk of severe illness due to nH1N1 was similar to seasonal influenza, enabling NYC to emphasize preventing severe morbidity rather than employing aggressive community mitigation measures.
The public health response to an emerging influenza pandemic, particularly whether to initiate aggressive community mitigation strategies such as school closure, depends in part on the severity of illness caused by the potentially pandemic strain: whether it has a more severe disease rate or higher mortality than usually seen with seasonal influenza
In the United States, a key measure for categorizing the potential severity of a pandemic strain is the case-fatality rate (CFR) among those infected
Population-based telephone surveys can be a useful way to quickly assess and monitor the prevalence and distribution of influenza-like illness (ILI) in the community. However, they do not distinguish between influenza and other causes of ILI (e.g., respiratory syncytial virus, rhinoviruses, coronaviruses, parainfluenza viruses) and their use to rapidly estimate the prevalence of influenza is challenging.
On April 24, 2009, New York City (NYC) became the third geographic area of the US in one week to document the presence of novel H1N1 (nH1N1), later declared pandemic influenza (H1N1) 2009
This paper presents our estimates of the prevalence of nH1N1 during the 50 days of peak circulation of nH1N1 in NYC, the methods used to derive them, and the resulting case-fatality, hospitalization and intensive care unit (ICU) admission rates. To our knowledge, NYC was the only area of the country to directly obtain population-level ILI or nH1N1 prevalence during the initial spring 2009 wave of the pandemic.
Methods used included serial population surveys to estimate ILI prevalence for the time period May 1–June 19, 2009, use of two separate methods to determine and then discount estimates for background ILI in order to estimate nH1N1 prevalence, and calculation of case-fatality, intensive care unit (ICU) admission and hospitalization rates using data from surveillance for deaths, ICU admissions and hospitalizations during this time period for numerators.
The investigation of this novel strain of influenza in April/May, 2009, including the population-based telephone surveys, was deemed public health practice and not human subjects research by both the General Counsel and the Institutional Review Board Chair of the New York City Department of Health and Mental Hygiene, and therefore did not require Institutional Review Board review.
DOHMH employed rapid telephone polling methods typically used in public opinion research to assess ILI prevalence among both adults and children during two overlapping periods in May and June 2009. We conducted two polls of approximately 1,000 households each, between May 20 and May 27, and between June 15 and June 19, 2009. This size sample was adequate to generate a reliable citywide prevalence estimate yet still be conducted quickly. Household samples of 1000 would typically yield data on more than 2500 persons. The predicted 95% margin of error around an estimated prevalence of 50% in such a sample is less than +/− 2%, even after adjusting for non-response weighting. We used a random-digit dialing telephone sampling methodology to obtain data from a random sample of residential households in NYC. A nonrandom adult from each household was asked to provide information on all household members. Interviews lasted 5 minutes and were conducted in both English and Spanish. Sampled numbers were dialed between five and six times to contact and interview a household, or until the sampled number was determined to be non-working. For the first survey, the Council of American Survey Research Organizations (CASRO) response rate 3 (RR3) was 8.4% with a cooperation rate of 31.2%
For each survey, the analysis dataset contained a record for every enumerated household member. Household members were linked by a household ID. Data were weighted to population estimates from the 2007 American Community Survey (ACS)
ILI was defined as having fever and either cough or sore throat during the specified time period. In the first survey, respondents were asked whether they or other household members had experienced fever and either cough or sore throat between May first and the date of the interview (May 20–27). Information on household members was recorded by age group. In the second survey, the same procedures were used, but the time frame was changed to the past 30 days (May 15–19 to June 15–19).
The surveys included two overlapping time periods with different estimates of ILI prevalence. To combine the surveys, an estimate was needed of the proportion of ILI cases reported in each survey that occurred in the 13 day overlapping time period, May 15–27. Ideally, this should reflect the underlying epidemic curve for ILI in NYC during this time. To approximate this, we used emergency department (ED) visit data from our syndromic surveillance system. This system includes daily information on 92% of all hospital ED visits in NYC. Chief complaints are used to classify visits into syndrome categories, including an ILI category that utilizes a definition similar to the survey question, but which can also include the word “flu”
For each survey, the number of ED ILI visits occurring during the non-overlap and overlap periods was calculated by age, sex, and borough. Assuming that the ILI prevalence reported in the survey followed the same distribution as the ED visits (
Rate is number of emergency department (ED) visits for influenza-like illness (ILI) per 100,000 age-group specific population. Survey 1 was conducted from May 21–27 and measured ILI from May 1–27. Survey 2 was conducted June 15–19 and measured ILI from May 15 to June 19. The overlap period is from May 15–May 27.
Two approaches were used to estimate background ILI activity expected in the absence of nH1N1. Each approach produced overall and age group-specific estimates of background ILI activity. Each estimate of background ILI activity was applied to the combined survey data point prevalence to produce estimates of the percentage of NYC residents affected by nH1N1 from May 1 through June 19, 2009. Ninety-five percent confidence limits, adjusted for the complex survey design, were calculated around the adjusted point estimates.
During the reporting period of the May survey, results from ongoing enhanced surveillance of laboratory-confirmed hospitalized nH1N1 cases and of emergency department visits for ILI suggested that community transmission of nH1N1 was occurring primarily in certain parts of the city. Most cases were coming from the area in Queens surrounding the location of the initial high school outbreak, and from selected parts of Brooklyn. Assuming no geographic variability in the ratios of nH1N1 hospitalizations to nH1N1 prevalence and of nH1N1 prevalence to ILI prevalence, we hypothesized that ILI prevalence in the less affected areas would be largely attributable to causes other than nH1N1. Following this logic, we defined background ILI prevalence as the average prevalence from the initially less affected boroughs, Bronx, Manhattan and Staten Island from the first survey, overall and by age group, and assumed the same background rate for the entire survey period. The rates for the first survey period were calculated based on the distribution by date of responses to the May 20–27 survey, averaging approximately 23.1 days from May 1 and, thus, giving a 23.1 day prevalence. This estimate was then expanded to the full 50 day period. We then subtracted our estimates of background ILI from the combined ILI prevalence estimates to produce estimates of nH1N1 prevalence for the period May 1–June 19, 2009.
The second method used data from the ED syndromic surveillance system to adjust for background ILI. Using data from EDs reporting consistently over the preceding 5 years, we constructed an age group-specific Serfling model, similar to what is used for estimating excess mortality due to influenza.
During this time period, active surveillance for the first 3 weeks followed by enhanced surveillance was conducted for deaths through all NYC hospitals and the medical examiner's office, including nH1N1 specific testing. For most of the period, enhanced surveillance for hospitalizations and ICU admissions due to nH1N1 was also conducted, including nH1N1 specific testing on persons hospitalized who tested influenza A positive and on all persons admitted to the ICU with acute respiratory illness, including ILI. Details of the surveillance efforts are described elsewhere
Two sets of case-fatality, hospitalization and ICU admission rates were calculated using numerators obtained through population-based nH1N1 death and hospitalization surveillance and each of the two sets of denominators of persons estimated to have had ILI due to nH1N1 from the population surveys after adjustment for background ILI. Overall and age group-specific rates were generated.
ILI prevalence estimates and confidence intervals were produced using SAS-callable SUDAAN Statistical Methods software to adjust standard errors for complex survey design. Compound estimates such as the combined survey estimate of nH1N1 prevalence attributable ILI were generated using Monte Carlo methods with the R statistical package.
Survey 1 |
Survey 2 |
Combined Estimate May 1–June 19 | |||||||
Number with ILI |
Percent with ILI | 95% CI | Number with ILI |
Percent with ILI | 95% CI | Number with ILI |
Percent with ILI | 95% CI | |
Total | 576,000 | 6.9 | 6.0–7.9 | 1,007,000 | 12.0 | 10.0–14.6 | 1,318,000 | 15.8 | 13.2–19.0 |
Sex | |||||||||
Male | 266,00 | 6.7 | 5.3–8.5 | 402,000 | 10.1 | 7.5–13.3 | 540,000 | 13.5 | 10.2–18.0 |
Female | 314,00 | 7.2 | 6.0–8.7 | 608,000 | 13.9 | 10.7–17.6 | 764,000 | 17.5 | 13.6–22.1 |
Age Group | |||||||||
0–17 | 224,000 | 11.7 | 9.2–14.7 | 405,000 | 21.1 | 15.9–26.7 | 510,000 | 26.6 | 20.1–34.0 |
18–64 | 306,000 | 5.7 | 4.6–6.8 | 544,000 | 10.1 | 7.6–13.3 | 717,000 | 13.2 | 10.2–17.2 |
65+ | 45,000 | 4.3 | 3.0–6.3 | 59,000 | 5.7 | 4.1–7.1 | 91,000 | 8.8 | 6.2–12.6 |
Borough | |||||||||
Bronx | 50,000 | 3.6 | 2.2–5.7 | 127,000 | 9.1 | 5.8–13.4 | 155,000 | 11.1 | 7.0–16.6 |
Brooklyn | 230,000 | 9.0 | 7.2–11.4 | 335,000 | 13.1 | 10.1–17.1 | 452,000 | 17.7 | 13.7–23.4 |
Manhattan | 61,000 | 3.7 | 2.3–5.8 | 184,000 | 11.3 | 7.8–15.5 | 217,000 | 13.3 | 9.1–18.6 |
Queens | 217,000 | 9.4 | 7.0–12.7 | 280,000 | 12.2 | 8.2–18.6 | 383,000 | 16.7 | 11.6–25.4 |
Staten Island | 20,000 | 4.2 | 2.3–7.4 | 87,000 | 17.9 | 7.3–34.8 | 98,000 | 20.1 | 8.6–38.9 |
*Survey 1 conducted May 21–27, covering time period May 1–27. Survey 2 conducted June 15–19, covering time period May 15–June 19.
**Number with influenza-like illness (ILI) calculated by multiplying the group-specific 2007 population estimates by the percent with ILI and rounding to the nearest 1000.
Total ILI prevalence during the May 1–May 27 period was 6.9%. Prevalence was considerably higher in children (11.7%) than in adults (5.7%) or older adults (4.3%). There was geographic variability in ILI prevalence during this period. Prevalence was high in Brooklyn (9.0%) and Queens (9.4%) and lower in the three boroughs chosen to represent background ILI, Bronx (3.6%), Manhattan (3.7%) and Staten Island (4.2%).
Total ILI prevalence for the period May 15, 2009–June 19, 2009 was 12.0%, with age-related differences ranging from 21.1% in children 0–17 years to 5.7% in those 65 years and older. No variability in ILI prevalence by borough was observed during this time period.
After combining data from the two surveys, the estimated overall ILI prevalence was 15.8% (estimated N = 1,318,000 NYC residents), with age-specific estimates ranging from 26.6% among children ages 0–17 years, to 13.2% among adults 18–64 years and 8.8% among adults ≥65 years.
The results of each approach after adjusting for background ILI to obtain estimates of nH1N1 prevalence are presented in
Combined Data May 1–June 19 | Adjustment Method 1 |
Adjustment Method 2 |
|||||
Estimated ILI Percent Prevalence (95% CI) | Estimated Percent Background ILI (95% CI) | Estimated Percent nH1N1 Prevalence (95% CI) | Estimated Number with nH1N1 (95% CI) | Estimated Percent Background ILI (95% CI) | Estimated Percent nH1N1 Prevalence (95% CI) | Estimated Number with nH1N1 (95% CI) | |
NYC | 15.8 (13.2–19.0) | 8.0 (5.9–10.9) | 7.8 (4.4–10.5) | 639,000 (367,000–880,000) | 3.6 (3.1–4.3) | 12.2 (10.1–14.6) | 1,017,000 (848,000–1,231,000) |
Age-Group | |||||||
0–17 years | 26.6 (20.1–34.0) | 13.5 (8.4–21.5) | 13.1 (4.5–20.2) | 250,000 (87,000–388,000) | 6.6 (5.0–8.5) | 20.0 (15.1–25.5) | 383,000 (290,000–488,000) |
18–64 years | 13.2 (10.2–17.2) | 6.6 (4.4–9.9) | 6.6 (2.6–9.4) | 355,000 (156,000–548,000) | 2.5 (2.0–3.2) | 10.8 (8.3–14.0) | 582,000 (446,000–758,000) |
65+ years | 8.8 (6.2–12.6) | 5.6 (3.0–9.8) | 3.2 (0.0–6.5) | 34,000 (0–68,000) | 3.7 (2.6–5.4) | 5.1 (3.6–7.2) | 53,000 (37,000–75,000) |
*Adjustment Method 1 uses survey 1 data from the less affected boroughs to estimate background ILI.
**Adjustment Method 2 uses emergency department visit data for ILI from 2004–2008 to estimate background ILI.
The two methods gave nH1N1 point-prevalence estimates that were approximately 1.6-fold different. Method 1 resulted in an overall nH1N1 prevalence of 7.8%, with age-specific prevalence of 13.1% for children <18 years, 6.6% for adults 18–64 years, and 3.2% for those ≥65 years. By contrast, method 2 resulted in nH1N1 prevalence of 12.2% overall and 20.0%, 10.8%, and 5.1% in children, adults, and older adults, respectively.
Adjustment Method 1 | Adjustment Method 2 | ||||
No. Cases | No. persons nH1N1 |
Rate/1000 persons nH1N1 |
No. persons nH1N1 |
Rate/1000 persons nH1N1 |
|
Fatalities | |||||
All | 55 | 639,000 | 0.086 | 1,017,000 | 0.054 |
0–17 years | 3 | 250,000 | 0.012 | 383,000 | 0.008 |
18–64 years | 47 | 355,000 | 0.132 | 582,000 | 0.081 |
65+ years | 5 | 34,000 | 0.147 | 53,000 | 0.094 |
Hospitalizations | |||||
All | 859 | 639,000 | 1.34 | 1,017,000 | 0.84 |
0–17 years | 377 | 250,000 | 1.51 | 383,000 | 0.98 |
18–64 years | 440 | 355,000 | 1.24 | 582,000 | 0.76 |
65+ years | 42 | 34,000 | 1.24 | 53,000 | 0.79 |
ICU admissions | |||||
All | 214 | 639,000 | 0.335 | 1,017,000 | 0.210 |
0–17 years | 81 | 250,000 | 0.324 | 383,000 | 0.211 |
18–64 years | 122 | 355,000 | 0.344 | 582,000 | 0.210 |
65+ years | 11 | 34,000 | 0.323 | 53,000 | 0.208 |
*Point estimate of number of persons with nH1N1 from
**[No. cases]/[No. persons with nH1N1]×1,000.
The overall hospitalization rate ranged from 0.84 to 1.34 per 1,000 persons with ILI due to nH1N1. Children 0–17 years were at slightly higher risk for hospitalization than those who were older (RR = 1.22, 95% CI 1.06–1.39, p<0.01, adjustment method 1; RR = 1.30, 95% CI 1.13–1.48, p<0.001, adjustment method 2).
The overall ICU admission rate ranged from 0.21 to 0.34 per 1,000 persons with ILI due to nH1N1. There was no variation in ICU admission rates by age. However, the percentage of hospitalized cases who were admitted to the ICU did vary slightly by age. Adults 18 years and older were more likely than children 0–17 years to be admitted to the ICU (27.6% vs 21.5%, RR = 1.28, 95% CI 1.01–1.63).
It is critical to assess the severity of a potentially pandemic strain of influenza as soon as possible after it is recognized in order to inform the public and to guide public health response. Defining the risk of death following infection with a pandemic strain enables categorization of the potential severity of the pandemic, with a CFR of <1 death per 1000 persons infected being the criteria for the lowest severity, Category 1 pandemic strain
Measuring influenza CFRs in “real-time” to inform public health efforts is a challenge that was acknowledged at the beginning of this pandemic
These methods have produced widely different measures of CFR, hospitalization and ICU admission rates due to pandemic H1N1 influenza in the US. The CFR ranges from a potentially low estimate of 0.05–0.09 per thousand based on our population survey data and 55 deaths, to a potentially high estimate of 0.48 to 5.1 per 1000 based on use of multipliers tied to 788 medically attended confirmed infections, 25 hospitalizations and 4 deaths in Milwaukee
Why were there such large differences between the two ways of estimating severity, especially since they have different implications for hospital preparedness? Is one method potentially more accurate than the other? We believe that the method used in NYC is more likely to produce an accurate measure of CFR simply because it is only dependent on two measures: population-level infection and deaths. The multiplier method is dependent on more measures and includes projection from the number of people diagnosed to the number seen for ILI and from the number seen to the number who were symptomatic. In this case, two additional factors could play a role: the numbers of confirmed cases and hospitalizations used in the multiplier model were small (788 and 25) and the data used to project from confirmed cases to estimate the population affected were obtained from studies done elsewhere and in special settings during the H1N1 pandemic (Chicago and Delaware) or from community surveys done when seasonal influenza was circulating
We believe the NYC data, while still reflecting a range of estimates, to be fairly accurate for at least three reasons. First, numerator data were based on enhanced death surveillance including deaths referred to the medical examiner with specific testing for nH1N1 of all suspect deaths. Outpatient deaths were able to be identified and included. Second, despite low response rates on the survey, our ILI and nH1N1 prevalence data are consistent with data from other sources. The first survey showed a strong age-specific gradient with much higher prevalence in areas of NYC with initial amplification of nH1N1 and the data from the second survey were more uniform and much higher. These findings were consistent with hospitalization and ED syndromic surveillance data. In addition, the methods for adjustment for ILI likely produced artificially low and high estimates of nH1N1 prevalence, estimates that encompassed the actual prevalence. Assuming the baseline ILI prevalence in the absence of nH1N1 was the measured ILI in the first survey in boroughs with minimal nH1N1 activity based on hospitalization and ED syndromic surveillance data, we likely overestimated the background ILI rate and, correspondingly, had low nH1N1 prevalence estimates. Halfway through the first survey time period, it was clear from ED data that visits for ILI were increasing in those boroughs, a sign of nH1N1 activity spreading to them. Thus, the background ILI rates likely included nH1N1-related ILI. On the other hand, the ED syndromic surveillance adjustment method likely overestimated total ILI rates and, correspondingly, overestimated nH1N1 rates. In NYC, as in many other places in the US, people with ILI appeared to be more likely to go to the ED than normally would have in hopes of getting tested for nH1N1. Third, NYC clearly had a high nH1N1 prevalence during the first pandemic wave, one of the highest in the country based on death reports (25% of US reported nH1N1 deaths as of July 2, 2009)
Except for those 65 years and older, our estimates of age group-specific CFR for nH1N1 are almost identical to those derived from mortality data for seasonal influenza from 1990–1999, the data that forms the basis for the widely cited statistic that seasonal influenza causes an estimated 36,000 deaths per year in the US
During the spring wave of nH1N1, it was noted in several journal articles that it is very difficult to measure case-fatality rates early in a pandemic
The major limitations of this study include the low survey response rate and the inability to measure nH1N1 prevalence directly. The low survey response rate was in part a predictable and necessary consequence of the rapidity with which the surveys were conducted, a limitation that can be expected whenever rapid polling methods are used. Additional survey-related limitations are those associated with any telephone survey, specifically recall, self-report and the potential for those with land line telephones to be different than those in the rest of the population. There are also limitations associated with conducting death surveillance and hospitalization surveillance. If a clinician does not think that a death could be influenza-related and/or fails to conduct testing or report to either public health authorities or the medical examiner, it will not be recognized and counted. Hospitalization surveillance relied in part on initial screening testing with an insensitive rapid antigen test for influenza A, and upon both clinician recognition of influenza and reporting in response to frequent telephone prompts. For ICU surveillance, however, only the clinical recognition of influenza and reporting was a limitation: nH1N1 specific testing was offered to all with acute respiratory illness, including ILI. These limitations all result in under measurement of fatalities and hospitalizations. The strengths of this approach include the rapid availability of the data, and the ability to capture data on children as well as adults. Of note, as of June 2010, no nH1N1 seroprevalence data on NYC residents following the Spring 2009 outbreak has become available to assess the accuracy of the prevalence estimates presented in this paper.
We would like to thank Stephen Immerwahr for his help in designing and managing the surveys, Marci Layton and Thomas Farley for their review and encouragement, the many members of the NYC Swine Flu Investigation Team for their intensive surveillance efforts and Global Communications for carrying out data collection.