In developing countries, rotavirus-associated gastroenteritis is a leading cause of morbidity and mortality in early childhood. The burden of rotavirus gastroenteritis is highest among very young children and decreases rapidly thereafter. In temperate areas, seasonality of rotavirus infection with a single winter peak is well documented. In tropical climates, seasonality is less defined, with occasional increase noticed during cold and dry periods [1, 2].

The Indian ‘National Rotavirus Surveillance Network’ (NRSN) was initiated in 2005 by the Indian Council of Medical Research (ICMR) as a collaborative task-force study to generate nationally representative data on rotavirus burden and its molecular epidemiology in India. The multi-site, hospital- based surveillance has been carried out in two rounds (Round 1: 2005-2009 and Round 2: 2012-2016). The data from 2005-2009 has been previously analyzed to report association of rotavirus positivity with age, gender and season as well as the distribution of specific genotypes [3,4].

With multi-site data sets, the application of appropriate statistical measures to more meaningfully interpret national level data is important. Such analyses facilitate interpretation of variations by years, by regions, by population characteristics like age and gender, and by seasonality. These may be useful in planning and implementing appropriate prevention and control strategies in different parts of the country. We used two new methodological approaches namely ‘Adjusted Proportion’ and ‘Symmetrized Index’ (adjusted variation) to assess temporal, geographical and demographical variations in disease patterns, and have illustrated the application of these measures in the context of rotavirus gastroenteritis in India.

In addition to calculating the prevalence of rotavirus positivity (a measure of rotavirus disease burden), to understand the disease pattern in different regions over time and across different groups (levels of aggregations), we calculated the following two statistical measures:

Rotavirus prevalence for the period 2005-2009 ranged from 37% to 44% with maximum, minimum and average rotavirus positivity varying across the regions both annually (Web Fig. 1; Table I) and seasons during the survey period (Table I). The minimum rotavirus prevalence (13.8%) was observed in the third season of 2007 and the maximum prevalence (88.9%) was observed in first season of 2006. The annual average prevalence ranged from 36% (95% CI 29.5- 43.6) to 44% (95% CI 39.6 – 48.9). In the Eastern region during June-Aug of 2007, rotavirus positivity was nil and maximum prevalence of 81% was observed in September – November of 2006 (data not shown). In the Western region, lowest prevalence of 5% was seen during June-August of 2007 and maximum prevalence of 80% was observed in September-November of the same year. In the Southern region, lowest and highest prevalences were 19% (March -May) and 68% (December-February) of 2008, respectively. In the Northern region, lowest positivity was seen in September -November (13%) of 2008 whereas the highest prevalence was observed in December- February (87%) of 2009 (data not shown).

TABLE I Annual and Seasonal Prevalence of Rotavirus Disease in India

Using adjusted proportions, differences in rotavirus positivity for various domains (annual, seasons, age group etc.) adjusted for reference points (domain prevalence/ mean) with 95% CI are shown in Table II. Deviation from the annual prevalence (adjusted proportion of >1.0) was observed for the season December-February and September – November in all the regions (Table II). A similar phenomenon was observed for the age groups of <12 months and 12-23 months and in the Eastern and Northern regions.

TABLE II	Adjusted Proportions for Rotavirus Positivity Across Domains

Symmetrized Index (SI) varied from 4% to 52% during the surveillance period across regions when the rotavirus positivity was accounted for the seasonal periodicity (Table I). For the entire study period, the calculated estimate SI was 67%. When accounting for annual periodicity the SI ranged from 38% to 64%, highlighting the high rotaviral disease burden. Web Fig. 2 depicts the annual-seasonal and regional distribution of SI. Analysis of variance performed on SI revealed that there was no variation among SI accounting for seasons and age of children during the study period (data not shown).

In the present surveillance data, although hospitalization with rotavirus gastroenteritis occurred throughout the year, more hospitalizations were seen during months of December to February. Fluctuations in rotavirus disease burden can be detected and modeled using a range of statistical tools [6]. Measurements in terms of rates, percentages, proportions etc. can be calculated for different regions or for domains (levels of aggregations) that are mutually exclusive and exhaustive. Patterns become evident when such quantitative measures are compared between various domains, but such summarizations involve loss of information [7]. To compensate for this loss, normalization of data or adjusting for average prevalence or symmetrization of extreme observations in prevalence will result in refinement in measurements [8].In the present analysis, comparison of estimates derived by the usual prevalence measurement and the symmetrized index revealed that symmetrization resulted in a higher estimate of 67%. Similar increase or decrease in estimates was observed when SI and prevalence was analyzed for different seasons and regions. By accounting for the difference in extreme observations, the analysis shows that the symmmetrized index will provide more realistic estimates of rotavirus disease variability. This "folded fraction" is particularly useful in settings where the range is large and outlier observations are more.

The graphical depiction of prevalence in this paper aids understanding of the distribution of the disease burden. Adjusted proportion analysis explains the variations among differences in annual or seasonal and other such data categorization and calculation of confidence intervals for adjusted proportions adds value to such estimates. This ratio reduces noise from the settings in which little or no rotavirus was observed. It also gives an indication of greater deviation from the average whenever this ratio is greater than one. Usually calculation of confidence intervals is done for simple proportions. The recent development of calculating confidence intervals for ratio of proportions allows application of the indirect test of significance for equality [5]. Using adjusted proportions, we have shown that one can easily identify regions / seasons where the deviation in rotaviral prevalence is more than the average. Therefore this measure can be used to generate pattern of rotaviral disease in the country showing the differences in domains over time.

In conclusion, the analysis using these newer and simple statistical measures has potential for application in surveillance of other major infectious diseases such as meningitis or pneumonias wherein evaluation of the load and pattern of disease burden in different regions over time can be carried out.

Acknowledgements:　 M Chiranjeevi, Technical Assistant, NRSN project team at NIE for support in statistical analysis. Dr Gagandeep Kang for reviewing and providing valuable inputs.

Contributors: SV: manuscript conceptualization and development, data analysis; CPGK: concept refinement, data interpretation and manuscript writing; SMM: concept refinement, data interpretation and manuscript writing. All authors approved the final manuscript.

Funding: Indian Council of Medical Research; Competing Interests: None