Death and Population Data

We obtained the monthly number of deaths caused by P&I and AC in Italy from 1969 (first available data year) through 2001 (most recent data year) from the Italian National Census Bureau, which records all causes of death reported on death certificates and classifies them according to the International Classification of Diseases (ICD). For our analysis, we considered only the underlying cause of death. To select P&I deaths, we used ICD-8 codes 480–486 and 470–474 for the years 1969–1979 and ICD-9 codes 480–486 and 487 for the years 1980–2001.

The geographic areas considered were the 3 administrative regions of Italy: northern Italy, central Italy, and southern Italy, as defined by the Italian National Census Bureau (20). Northern Italy comprises Piedmont, Lombardy, Autonomous Province of Trento, Autonomous Province of Bolzano, Val d’Aosta, Veneto, Friuli Venezia Giulia, Liguria, and Emilia Romagna. Central Italy comprises Tuscany, Umbria, Marche, and Lazio. Southern Italy comprises Abruzzo, Molise, Campania, Puglia, Basilicata, Calabria, Sicily, and Sardinia. For each year, we generated summary datasets of the monthly number of deaths from P&I and AC, stratified by age group (0–14, 15–44, 45–64, and >65 years) (9). We calculated the annual number of persons in each age group and the monthly number of deaths per 100,000 population for each age group and standardized these to 30.4-day months.

Virologic Surveillance

To determine which influenza viruses were circulating each season, we reviewed publications listing viral subtypes identified in Italian laboratories (21–23). For the most recent years (1999–2001), we obtained these data from the Italian National Influenza Center, which has performed virologic surveillance since 1999.

Statistical Analyses

To estimate age-specific excess deaths an indirect measure of death attributable to influenza from P&I and AC for the 32 influenza seasons, we applied a Serfling-type regression model to monthly time series of deaths (7,9). As described in previous studies (9,19), we removed the seasonal trend from the time-series data (de-trended) by fitting a smooth spline function to the average death rates in summer (June–August). Then, we applied a seasonal regression model to the de-trended series, excluding values for December–April, to model the expected mortality rates in the absence of influenza activity. Monthly mortality rates were calculated as the observed minus the predicted mortality rates for all epidemic months. We identified epidemic months by applying the above-mentioned procedure to deaths coded specifically as influenza (ICD-8 code 470–474 and ICD-9 code 487). We defined epidemic months as those winter months for which influenza-specific mortality rates exceeded the upper 95% confidence limit of the seasonal model.

Seasonal excess deaths were estimated as the sum of monthly excess deaths, after back-adjusting for the true month length and removing the spline transformation. The model was applied to P&I and AC mortality rates separately for each age group. We achieved an excellent fit for all age groups. All model terms included were statistically significant (p<0.0001), but additional terms were not (p>0.05).

Age Patterns and Geographic Trends

To determine whether variations in age structure biased the geographic comparisons, we generated excess mortality rates for each area and age group and standardized them on the basis of the age distribution of the Italian population in 2001 (the year of the most recent Italian census). This permitted a comparison of age-adjusted P&I and AC excess mortality rate estimates across areas.

To compensate for nondemographic differences among areas (e.g., differences in access to healthcare and in coding for cause-specific deaths) (24), we also calculated the percentage increase in mortality rates as the excess deaths divided by the baseline deaths in winter (expected deaths), for P&I and AC, separately. This measure has been used successfully in past research (7,8). To estimate correlations of influenza-related death across the 3 geographic regions, we calculated the pairwise Spearman correlation coefficients of seasonal estimates for the 32 years considered.

Geographic Trends, Synchrony, and Effect of Epidemic Seasons (All Ages)

For the 32 influenza seasons studied, excess deaths averaged 3 per 100,000 population (range 0–38) for P&I and 18 per 100,000 for AC (range 0–107). Influenza accounted for an estimated 57,243 deaths from P&I (average 1,789 per season) and 318,806 from AC (average 9,963 per season).

Figure 1

Figure 1. Monthly mortality rates from pneumonia and influenza and all causes for Italy, January 1969–December 2001. *Baseline mortality rates determined by Serfling model.

The highest number of excess deaths was found for the 1969–70 pandemic season; no measurable number of excess deaths was found for 5 seasons (1981–82, 1984–85, 1986–87, 1990–91, 2000–01) (Figure 1). The 27 seasons that had excess P&I and AC deaths had an average of 2.4 epidemic months per season (range 1–4). The influenza seasons with higher excess deaths tended to be characterized by a predominance of influenza A (H3N2) viruses (Figure 1). For these seasons, the average excess deaths from P&I and AC (4.5 and 23.4 per 100,000 population, respectively) was 4× higher than that for the 11 seasons in which influenza A (H1N1) or B viruses were predominant (0.8 for P&I and 7.4 for AC, per 100,000 population).

Figure 2

Figure 2. Correlation in the influenza epidemics for 31 influenza seasons (1970–2001), measured by excess deaths from pneumonia and influenza (A) and all causes (B). Excess mortality rates per 100,000 in southern and...

For the overall study period, the excess deaths per 100,000 population from AC was 15 for northern Italy, 14 for central Italy, and 22 for southern Italy; from P&I they were 4, 3, and 3, respectively (Appendix Figure 1). Also for these 32 years, no statistical differences among the 3 geographic areas were noted for excess deaths from P&I or AC (Kolmogorov-Smirnov test, p = 0.8 and p = 0.9, respectively). Patterns were similar with the percent increase in excess deaths from P&I and AC. The 95% confidence intervals for estimates for individual seasons were within 6% of given values (Table 1). When conducting this analysis for seasons in which influenza A (H3N2), A (H1N1), and B predominated, area-level differences were again not significant. The strong correlation of excess-death estimates in the 3 regions suggests a high level of synchrony in the amplitude of local influenza epidemics (number of excess deaths peaked in the same month in each region) across Italy (P&I, Spearman ρ = 0.88–0.93, p<0.0001; AC, Spearman ρ = 0.80–0.93, p<0.0001) (Figure 2; Appendix Figure 2).

Magnitude and Trends of Influenza-related Deaths during Pandemic and Epidemic Seasons, by Age

During epidemic seasons, most influenza-related deaths at the national level (84%) occurred in persons ≥65 years of age, for P&I and AC; by contrast, during the 1969–70 influenza A (H3N2) pandemic season, deaths markedly affected all age groups, especially the 45–64 group.

In Italy the proportion of excess deaths from AC in persons <65 years of age was 3-fold higher during the pandemic season than during all other epidemic seasons. In particular, when the pandemic season was compared with the season with the second highest number of deaths (1974–75), the number of influenza-related deaths was 7× higher for persons 0–14, 4× higher for persons 15–44 and 45–64, and 2× higher for persons ≥65 years of age. Similar results were obtained for all 3 geographic areas (Table 2). The number of excess deaths from AC during the influenza A (H3N2) pandemic season was 1- to 9-fold higher in Italy than in other European countries (France, England, and Wales), in North American countries (United States, Canada), and in Asian countries (Japan, Australia) (Table 3).