In 2004, the Epidemiology Department of the Ministry of Health of Morocco launched a year-round public sector syndromic surveillance system for ILI comprised of 412 primary health centers, with a catchment population of almost 12 million people. Sites report weekly ILI activity to the regional and central levels, where health officials aggregate the surveillance data. A case definition similar to the 1999 WHO ILI case definition recommended for public health surveillance, defined as “a sudden onset of fever, a temperature >38°C and cough or sore throat in the absence of another diagnosis” was used from 2004 to 2015 (15, 16). In 2015, Morocco adopted the updated WHO standard ILI case definition (5) developed in 2011 as “an acute respiratory illness with a measured temperature of ≥ 38°C and cough, with onset within the past 10 days” (17). Reporting includes the total number of ILI consultations aggregated by gender and age group, as well as total outpatient consultations. The proportion of ILI visits among all outpatient consultations is used as a proxy for influenza activity.
In 2007, the Moroccan National Influenza Center (NIC) began a virologic surveillance system in both ambulatory and hospital sites to complement the syndromic system and provide data on laboratory-confirmed influenza activity (18). After an interruption in data collection beginning in 2010, virologic surveillance was resumed in 8 sentinel sites in 2014. Specimens were collected and characterized between September and June. Enrolling patients from both out- and in-patient facilities allowed the integration of epidemiologic and virologic data representing the spectrum of illness from mild (ILI) to severe (e.g. severe acute respiratory infection or SARI) (17).
We used eleven seasons of syndromic surveillance data (2005/2006 to 2016/2017, excluding the 2009/2010 pandemic year from analysis as influenza activity was not reflective of a typical season); this was described elsewhere (19). We compared two methodologies for establishing seasonal baseline activity and epidemic thresholds. We also compared the calculated thresholds with the observed weeks for the start and end of the 2017/2018 season. Using three seasons of virologic ILI surveillance data (2014/2015 to 2016/2017), we used the MEM method to make calculations using the composite parameter recommended by WHO (20); this method estimates the proportion of laboratory-confirmed influenza ILI consultations among all outpatient consultations, or the product of weekly ILI consultations of total outpatient visits and weekly percentage of influenza-positive specimens among respiratory tests.
Methodology & statistical procedures
Overview of WHO and MEM methods
The methods discussed in order to standardize country information on influenza activity, have raised basic concepts summarized in table 1.
The WHO method
The 2012 WHO Global Epidemiological Surveillance Standards for Influenza (WHO Manual) (5) included a simple method to establish an average epidemic curve to identify the beginning of the influenza season using national influenza surveillance data. This method characterizes the intensity of influenza activity each year and may be used to describe the seasonality of influenza virus circulation. Using ILI as a proxy for influenza virologic activity (21, 22), we used weekly proportion of ILI among all outpatient consultations as our indicator of influenza activity.
With this method, we were able to produce an average epidemic curve. Using data from the average epidemic curve, we used statistical measures of variance to establish an alert threshold.
We determined the flat baseline for expected influenza activity throughout the year in order to develop an indicator for the onset of influenza season (seasonal threshold). Sustained influenza activity (i.e., three consecutive weeks) above this baseline indicated the start of the influenza season or the epidemic period (5). In the final step, moderate, high, and extraordinary intensity thresholds were estimated as described in the WHO Pandemic Influenza Severity Assessment manual (20), (Figure 1).
The Moving Epidemic Method
The Moving Epidemic Method (MEM) (25-30) is an alternative tool developed to help model influenza epidemics also using retrospective national surveillance data. It may be described as a combination rate-difference model that uses cumulative differences in rates to determine epidemic periods and intensity of activity (29, 30).
Using the free software R for statistical computing and graphics (27) and its open source user interface RStudio (28), we uploaded our surveillance data via the MEM application (25), and fit the model using three steps. We first visually compared activity over the eleven seasons in order to compare the timing of peak activity and activity trends across seasons. The MEM procedure has three main steps: First, the length, start and the end of the annual epidemics are determined, splitting the season in three periods: a pre-epidemic, an epidemic and a post-epidemic period (29, 30). In the second step, we built the model by using retrospective data from all eleven seasons. The MEM app calculated the pre-epidemic threshold that marks the start of the epidemic period (analogous to the seasonal threshold in the WHO method). In the third step, medium, high, and very high intensity thresholds were estimated (Table 2). Using the app, we produced graphs of each season showing the pre-epidemic, epidemic and post-epidemic periods (Figure 2). In addition, as the assumption that ILI activity is reflecting influenza virus circulation has limitations, we created a second seasonal threshold with this methodology using the composite parameter recommended by WHO for three seasons of virologic ILI surveillance (Figure 3).
Lastly, we calculated indicators of performance of the app to detect epidemics, using values from the model for sensitivity, specificity, positive predictive value, negative predictive value, percent agreement and the Matthew correlation coefficient (Table 3). The application allowed us to optimize the model by searching the optimum slope of the MAP curve to optimize the goodness-of-fit of the model for detecting epidemics.
The MEM app calculates goodness-of-fit indicators in an iterative process using a cross- validation procedure (29). True positives (TP) were then defined as values of epidemic period above the threshold, true negatives (TN) as values of the non-epidemic period below the threshold, false positives (FP) as values of the non-epidemic period above the threshold and false negatives (FN) as values of epidemic period below the threshold. The process was repeated for each season in the dataset and all TP, TN, FP and FN were pooled. To measure the performance of the threshold, the following statistics and definitions were used (29):
- Sensitivity: The number of epidemic weeks above the pre-epidemic threshold and above the post-epidemic threshold divided by the number of epidemic weeks (epidemic length).
- Specificity: The number of non-epidemic weeks below the pre-epidemic threshold and below the post-epidemic threshold divided by the number of non-epidemic weeks.
- Positive predictive value (PPV): The number of epidemic weeks above the threshold divided by the number of weeks above the threshold.
- Negative predictive value (NPV): The number of non-epidemic weeks below the threshold divided by the number of weeks below the threshold.
The ILI sentinel surveillance system is a public health activity organized by the Ministry of Health of Morocco. Personally identifiable data is excluded from this surveillance system; as a result, no request for authorization from the National Ethics Committees was required. Indeed, the Royal Dahir N°1-15-110 dated August 4, 2015, promulgating the law N°28-13 relating to the protection of persons participating in biomedical research, provides for special provisions for non-interventional or observational researches as stipulated in its articles 2 and 26.