The implementation of a surveillance system monitoring respiratory diseases at 9 Berlin hospitals initially in the context of a hospital-based vaccine effectiveness study was feasible and delivered important insights into the distribution of cases over time and according to different CD. A large proportion of pH1N1 positives was not captured by the SARI case definition as recommended by WHO and ECDC (CD4), with alternative case definitions having a better combination of sensitivity and specificity (Table ).
As expected, the incidence of both CD1b and CD4-cases was subject to the influence of seasonality. The proportion of cases fulfilling the narrowest CD4 definition was highest during study period 1 (winter), a period with substantial influenza virus circulation, and significantly lower thereafter. This was also reflected in the higher proportion of RI patients admitted to ICU during period 1. These findings are in keeping with a seasonal distribution of RI as described previously [18
]. Our results suggest that the spectrum of pathogens that circulate during winter causes a higher proportion of severe cases than the pathogen mix prevailing in spring and summer. Besides influenza, many respiratory pathogens exhibit seasonal patterns with peaks during the winter, e.g. adeno- or coronavirus infections [21
], but our data cannot provide insight as to which individual pathogens might cause increased morbidity. In our study almost a third of all RI-patients was positive for influenza during study period 1, and other studies have shown that influenza viruses are among the most common pathogens leading to hospitalization [21
] for respiratory disease or cardio-respiratory failure in the winter season [24
]. Nonetheless, a significant proportion of respiratory disease is apparently caused by other pathogens; even if possible underestimation of the proportion of influenza due to late swabbing in our study is taken into account.
We found that 20% of all hospitalized pH1N1-positive patients required ICU treatment, a number within the range of previously reported figures for pH1N1 illness [25
]. The higher risk of admission to ICU, mechanical ventilation and death in CD4 compared to CD1b cases was not statistically significant; this may have been due to insufficient power. Hospital stay was significantly longer in CD4 than in CD1b-cases.
The definition of CD4 (identical to the WHO/ECDC SARI case definition) captured only 55% of all hospitalized pH1N1-positive cases, and 75% of the pH1N1-cases requiring ICU management. Although a higher proportion of CD4 than CD1b-cases was pH1N1-positive, this difference was not statistically significant in either study period. Thus, we compared sensitivity and specificity of different case definitions to explore the ability of varying combinations of symptoms to capture PCR-confirmed pH1N1-influenza (gold standard) within our study population of hospitalized RI-patients. CD2, which permitted other systemic symptoms as an alternative to fever and did not require shortness of breath (SOB), had the highest sensitivity. The sensitivity of CD3, which differed from CD2 only in the strict requirement for fever, was only slightly lower. CD4 (SARI) had the lowest sensitivity. The trade-off for high sensitivity was a low specificity, which applied in reverse order to the tested case definitions. By comparing the areas under the ROC curves we took both measures -sensitivity and specificity- into account and found comparable and most favourable results for CD2 and CD3, at least during the period with significant influenza circulation. However, although the area under the ROC was highest for CD4, the difference to CD2 and CD3 was not statistically significant, likely due to overall low case numbers related to the rather late start of our study towards the end of the pandemic wave. The optimal syndromic case definition for respiratory infections to monitor severe (hospitalized) influenza cases has been subject to repeated discussions as, for instance, the 2011 update to the ILI (CD3) case definition by WHO Europe highlights: sore throat as a symptom was dropped and a history of fever as an additional qualifying symptom added [16
]. We were not able to assess the latter case definition as information on the presence of sore throat was not documented separately from cough in our study. However, our findings suggest that the broader case definitions offer advantages for surveillance of hospitalized influenza cases over the current SARI case definition as suggested by WHO/ECDC.
Hospitals showed a significant heterogeneity with respect to the proportions of patients admitted with RI. This suggests multiple study sites should be carefully selected to ensure representative surveillance of respiratory disease. Since we were interested in the trend of RI on the level of Berlin's population rather than single hospitals we did not stratify by hospital location.
Our study had some important limitations. The link to the vaccine effectiveness study led to a rather late start of our hospital-based surveillance, 2-3 weeks after the peak of the pH1N1-pandemic in Berlin and just a few weeks before it subsided. This resulted in the inclusion of a rather small number of pH1N1-positive patients. More comprehensive results would also have been possible if children and patients older than 65 years had been included in the study. Furthermore, the study ended just at the start of the 2010/2011 influenza season. Nonetheless, conduction of SARI-surveillance was found to be feasible in the framework of a hospital-based vaccine effectiveness study and can be considered in future pandemic situations if early implementation is possible. At least in our setting, the presence of trained study nurses was essential for the success of SARI-surveillance due to the extra work required to extract relevant information from the hospital information-system.
Almost a quarter (23%) of nasal and throat swabs was taken more than 7 days after symptom onset. This proportion differed by pH1N1-status: 5% among pH1N1-positives versus 24% among negatives (p
= 0.052), suggesting potential misclassification of pH1N1-positives as negative. However, other studies reported shedding of influenza virus for 6 to 9 days and, moreover, that shedding in hospitalized patients and with underlying illness appears to be longer than in households [27
]. In addition, we cannot rule out misclassification from inability to obtain or document symptoms in the hospital information-system. This could also be an explanation for the lower sensitivity of the CD4 case definition. Clinical information on underlying disease, which might serve to explain gender differences in regards of admission to ICU, was only available for patients included in the vaccine effectiveness study (results not shown), but was not available for all patients in the SARI-surveillance. Finally, our study was confined to one large city and is therefore not necessarily representative for other populations.