Figure 1: BlogViewer overview (detail).
BlogViewer helps analysts identify an epidemic's origin and spread from georeferenced microblogs. The software was developed in Processing, a Java-based visualization language for rapid prototyping, with the giCentre Utilities, a suite of code libraries for data visualization. It took 10-12 hours to produce; the first iteration (supporting spatial and temporal selection) was completed in 4-5 hours.
Whilst rapid prototyping requires programming ability, it is not beyond the scope of 'non-technical' individuals. Existing code libraries reduce the development effort, and an iterative approach encourages engagement with the data, as tools support analysis, and analysis leads to further requirements, in a symbiotic relationship.
The histogram shows the frequency of all 'ill' posts at multiple temporal resolutions (Figure 1); toggled bars show the frequency of first-time cases (Figure 2). The zoomable geospatial view can be overlaid with map points identifying the location of cases (Figure 1); a chi surface showing above or below-expectation cases for each grid cell (Figure 4); and a standard ellipse to summarise the distribution of cases (Figure 3). Selections can be made in space and time; the histogram, map points, standard ellipse, and scrollable list update in response. Selecting a post in the scrollable list both enlarges the post, and displays the poster's activity (previous and subsequent posts), with both flu-like symptoms and the selected post highlighted. Below the histogram, a weather summary indicates wind speed and direction (using arrow size and direction), and weather conditions (using a categorical colour scheme).
We believe Vastopolis experienced two epidemics:
Started around 6am, May 18. Evidenced by an increase in cases (taller histogram bars; Figure 2). Advancing a temporal selection suggests first 'ground zero' was in Downtown, around the Dome, City Hospital, and Convention Center. The chi surface supports this, and suggests the outbreak also affected Eastside (west of St Kirstin's Hospital; Figure 5). Suffers experienced flu-like symptoms.
Started around 2am, May 19. The chi surface (Figure 4), points map (Figure 2), and histogram (Figure 2; second peak in first-time cases, May 19) suggest this spatially and temporally distinct outbreak affected the river area south-west of the IS610 bridge (second 'ground zero'). Sufferers experienced stomach pain, diarrhoea, nausea, cramps. Poster histories (Video 00:50) suggested no overlap between the two symptom-groups.
We believe that the first epidemic was spread by airborne transmission, and was contained. We would not have advised the emergency services to deploy outside the affected area. This is because:
During and after the first epidemic, the wind direction was blowing towards the east or east-south-east (weather summary; Figure 1). The chi surface shows that cases were predominantly due east of the first 'ground zero' and affected 'central' Downtown and Eastside, west of St Kirstin's Hospital (green cells; Figure 5). In addition, there were below-expected cases to the north, in Uptown, and south, in 'lower' Downtown (purple cells; Figure 5). Consequently, we believe that the first epidemic was spread by airborne transmission. In addition, we suggest that the first epidemic followed a 'plume' pattern (narrow to the west, wide to the east) that may be the result of high-rise 'urban canons' found in central metropolitan areas, such as Downtown, opening out to low-rise areas, such as Eastside.
Although the first epidemic appeared to start simultaneously around the Dome, City Hospital, and Convention Center, we believe that the short distances from the Dome and Convention Center to the City Hospital, and the software's one hour minimum temporal resolution, over-emphasise the hospital. Indeed, when we compare all cases to first-time cases on the points map, we see that cases around hospitals probably reflected admissions: People fell ill, traveled to hospital, and continued to post whilst there (Video 00:30). Consequently, we believe that cases around the City Hospital were probably the result of 'walk-ins' from the immediate surrounding area.
When we select the area affected by the first epidemic in the geospatial view, toggle the standard ellipse, and advance a first-case temporal selection, we see that the first epidemic spread slowly from Downtown into Eastside, and that the concentration of cases decreased over time. This decrease is evident in the histogram: First-time cases exhibiting first epidemic symptoms (fever, chills, fatigue) declined after a peak between 6pm and 7:59pm on May 18. Indeed, the peak was slightly earlier, between 4pm and 5:59pm, in the affected area (histogram; Figure 3). Consequently, we believe that the first epidemic was contained.
We believe that the second epidemic was spread by waterborne transmission, and was probably contained. Whilst we would not have advised the emergency services to deploy outside the affected area, we would have given specific advice to the Vastopolis authorities (see below). This is because:
From the points map (Figure 2) it is clear that many cases were located around the river area south-west of the IS610 bridge: Indeed, the chi-surface suggests that these cases were above-expected (green cells; Figure 6). This evidence, together with the information that the river flows towards the south-west and supplies drinking water to the locality, suggests that the second epidemic was spread by waterborne transmission. Indeed, second epidemic symptoms (stomach pain, diarrhoea, nausea, cramps) are those of gastroenteritis, a condition spread through contaminated food and water.
When filtered by second epidemic symptoms, the frequency of cases peaked -- from zero -- between 2am and 3:59am on May 19; then diminished and rose to a lower peak between 8pm and 9:59pm on May 19; then diminished to zero from midnight, with several false positives throughout May 20. Summarising the distribution of cases using a standard ellipse (as above) we see that the concentration does not move downstream (Video 03:30). Consequently, we believe that the second epidemic was probably contained.
In the absence of additional georeferenced microblogs, we would have advised the Vastopolis authorities to contact their counterparts to the south-west, beyond the mapped area, to confirm that the second epidemic did indeed decline after May 20. Evidence from within the mapped area suggests this was likely, as the second epidemic diminished in concentration, even during the peak on May 19, to the south-west (green cells to south-west; Figure 6). In addition, we would have suggested the Vastopolis authorities protect the reservoirs to the east, given the terrorist threat to the city (Mini-Challenge 3).
Finally, population counts suggest that there is significant daily migration to and from Downtown and Westside, areas affected by the first and second epidemics respectively. However, this migration is not evidenced in either epidemics' spread. Indeed, the chi surface (Figure 4) shows that only hospitals outside the areas affected by both epidemics experienced above-expected cases, whilst the surrounding neighbourhoods experienced below-expected cases. (Interestingly, the exception is St George's Hospital in Uptown, which although outside an affected area experienced below-expected cases. However, there is evidence to suggest its medical staff were not admitting patients with first epidemic symptoms, rather administering pain-relief or antibiotics and sending them home directly.) Consequently, we believe neither epidemic was transmitted through person-to-person contact.
