Response volumes
A total of 34,144 distinct responses were collected between April 10 and July 29, 2020. Out of these, 1250 (3.7%) responses were positive for having any emergency symptoms and were subsequently referred to the next step in care management. A further 14,340 responses (42.0%) indicated the presence of other, non-emergency symptoms potentially related to COVID-19. Another 3112 (9.1%) responses indicated the presence of high-risk health conditions (e.g., heart failure, diabetes, age > = 70, auto-immune disease, immunotherapy) without potential COVID-19 symptoms being present. Figure 1 shows the number of responses to the self-assessment portal by week. The distribution suggests that the peak inquiries on the platform occurred between May 24 and June 28, 2020, with the highest volume occurring in the 3rd week of June.
COVID-19 indicator patterns over time
Responses to the portal questionnaire differentiated between the presence of any of a set of emergency symptoms (e.g., shortness of breath) that would require immediate medical attention and other symptoms that may be indicative of COVID-19 but were not considered to require emergency care. Figure 2 shows the numbers of portal responses with emergency symptoms present as well as the number of actual COVID-19 cases reported by government of Ontario in the postal code (FSA) ‘N’ over same time. Figure 3 shows the corresponding distributions of any potential COVID-19 symptoms (emergency or other) and confirmed cases in the region. In both plots, two y-axis scales are shown to allow comparisons between the trends in portal responses and COVID-19 cases confirmed by governement of Ontario.
Geographic response patterns
A total of 31,016 (99.6%) responses came from individuals who stated that they live in the FSA ‘N’ area, the primary catchment area for which the portal was set up. Figure 4 shows the geographical distribution of response volume, and the symptoms count distribution by FSA area. Because most responses came from the ‘N’ region, we limited our analysis to focus on this region. Within the ‘N’ FSA, the N2 region (see Online Appendix S1 for table of locations for each sub-region) recorded the highest absolute number of responses with 10,464 responses (Fig. 5), while the ‘N7’ region had the lowest absolute number with 358 responses during the same period. During the first 5 weeks of the data collection, the absolute number of responses from the ‘N6’ was the highest, but from the 6th week, the ‘N2’ region recorded the highest numbers per week. In the 11th week, the ‘N2’ FSA recorded an exponential increase in the number of responses (Fig. 6). This may be due to data error or a real surge in responses from this region in response to differential COVID-19 outbreak patterns, which will require further investigation to ascertain the origin.
To compute the comparative response levels between the FSAs, we used Statistics Canada’s 2016 census data6 to estimate the number of responses per 1000 residents in each postal code sub-region. FSA beginning with N6 had the highest per capita response, whereas those beginning with N7 had both the lowest per capita response (Fig. 7).
Symptoms reported by portal respondents
Figure 8 shows the trends in symptoms reported over time by type of symptom. All symptoms peaked in early May with a decline thereafter; however, a smaller, brief secondary peak was evident at the end of July. Sore throat was the most common symptom reported by respondents with 5880 of 31,016 non-emergency symptom respondents reporting they have the symptom either alone or in combination with other symptoms. The five most reported symptoms (Fig. 9) were sore throat (17.2%), headache (12.9%), fatigue (12.3%), digestive problems (12.2%) and cough (9.1%).
In terms of volume of the symptoms, respondents from FSA N2, N5 and N6 reported the most symptoms, while respondents from FSA N8 and N9 reported the most average symptom per respondents. However, further analysis showed that the distribution of symptoms within each FSA was largely similar (Fig. 10).
Symptom counts and clusters
There were regional variations in counts of symptoms reported in the Ontario Health West COVID-19 portal. The most common pattern was that respondents had no symptoms recorded; however, this is an artefact of two issues. First, if the respondent had any emergency symptoms, they were not asked any further questions about other symptoms or their health region. Instead, they were advised to seek immediate medical attention. Second, persons who had travelled to regions affected by COVID-19 or who had been in contact with persons with COVID-19 were not asked symptom-related questions. Both subgroups are present in the zero symptoms group.
On the other hand, up to 30 percent of respondents in some subregions reported three or more potential COVID-19 symptoms. There are regional variations evident in the symptom counts, but caution should be exercised in the interpretations of these differences because of the challenge in differentiating respondents who truly had no symptoms from those who were not asked the full set of questions about symptoms due to skip logic built into the portal.
Using Spearman rank sum correlation matrix, we examined patterns of associations between reported symptoms. The results suggest that fatigue tended to be accompanied by reports of chills, headaches, digestive symptoms, and sluggishness among children. Not surprisingly, fever and chills were associated with each other, as were congestion and runny nose. Following the implementation of a K-Mode machine learning clustering algorithm to the dataset to identify additional clusters, no other clusters were found beyond those evident from the correlational analyses in Table 1. It was not possible to conduct classification (supervised) machine learning modelling with the data. In part, machine learning performance was further limited by the lack of outcome or criterion measures (labels) that could be linked to symptoms at the person level.
Linkages to primary care
An important function of the COVID-19 self-assessment portal was to connect persons with potential symptoms to their primary health care provider when appropriate. Among 15,619 responses with any symptoms present (excluding emergency symptoms), 74.2% (11,589) had a primary health care provider that they were referred to for follow-up. These respondents were able to avoid emergency room visits and directed to contact the COVID-19 assessment center, because they could be appropriately served by primary care based on their response profile. About 17% of responses with symptoms had no primary health care provider; however, they were supported through access to an on-call provider.
Relationship between portal symptom reporting and sub-regional outbreaks
Following the initial analysis suggesting similar trends between number of respondents with symptoms on the portal self-assessment and actual number of COVID-19 cases in the region, we conducted further in-depth statistical analysis of this association. We standardized the number of portal symptoms and COVID-19 case by computing each number per 1000 population of the respective postal code. To ensure we covered for possible temporal variations in association, we computed up to 3 weeks lag and lead correlations between the portal symptoms and actual COVID-19 outbreaks.
For all the sub-regions, the number of people who reported each symptom were mostly negatively correlated with the number of COVID-19 cases. However, in a few sub-regions there were some significant positive correlations between number of symptoms reported through the portal and the actual number of COVID-19 cases reported to government of Ontario.
In regions with first two digits of the postal code, ‘N4’–‘N6’, the symptoms inconsistently showed significant positive correlation with actual number of COVID-19 cases for the region, with a 2–3 weeks lag time. This seems to suggest that in these sub-regions, portal symptoms reports started to rise about 2 or 3 weeks after the increase in actual COVID-19 cases reported to government of Ontario for each of the sub-regions. This suggests that individuals came to the portal in response to outbreaks happening in their sub-region.
Conversely, for sub-regions with the first two digits of postal code of ‘N3’, there were some positive correlations between some symptoms and actual COVID-19 cases, but with a lead time suggesting that people increasingly began to report such symptoms on the portal one to two weeks before COVID-19 cases began to rise in the area. Table 1 shows symptoms that were found to have positive lead or lag correlations with COVID-19 cases per sub-region. Some sub-regions, however, did not have any positive association with actual COVID-19 cases.