Demo data for Tableau data science demonstrations

Drug performance mining

This demo showcases Tableau's new regression modelling capabilities by identifying whether a particular medication performs above or below the expected value based on how patients rate the drug. Consider, for instance, this comparison of duloxetine (Cymbalta) and mirtazapine:

Both drugs have roughly the same number of users. However, Cymbalta is rated quite lower than mirtazapine, thus in relation to each other, Cymbalta is a better performer than mirtazapine.

COVID-19 geographical dynamics: the tale of three states

This demo showcases

• spatio-temporal splitting of a case-incidence time series,
• comparing a timespan average versus a rolling comparison of the preceding time segment (use the date range slider to set the index date, and the choropleth will indicate % change in TPR against the preceding 14-day window), and
• trend lines.

There's a detailed Wiki entry that explains the narrative behind this use case.

Spatial analysis of fatal drug poisonings, 1998-2016

This demo showcases

• spatial patterns: visualising spatiotemporal incidence patterns at a granular geographical level (US counties and the Federal District),
• time series trend detection: hovering over the individual counties shows the sparkline for the given county, and
• time series forecasting: shows a forecast at the 'tail' of established data, with an ambit of uncertainty (95% CI) based on a GLM.

Time series forecasting

This demo uses the data set on the global average temperature deviation in degrees Celsius to showcase predictive capabilities, in particular

• using an external time series prediction (using Prophet in Python),
• displaying actuals (blue), predicted (orange) and 95% CIs (green and teal, respectively), and
• the integration of seasonality (from monthly data) into the forecast.

Reliability and adverse events

This demo shows anomaly detection capabilities on the illustration of a (fictional!) data set simulating five side effects of a drug in three common patterns: constant-rate, constantly increasing rate and accumulative effects, where a previously unidentifiable part of the treatment cohort who are so susceptible exhibit the side effect after a given time in treatment. This illustrates

• using Tableau to identify anomalies, and
• using rolling calculations to identify the rapid spike in encephalopathy, isolated in time.

Survival

Typically, survival is visualised using the stepwise cumulative visualization. This is not always a useful way to see subcohort patterns. This use case utilises the data set by Haberman et al. to display what fraction of individuals who had surgery for breast cancer in a given year survived or did not survive past the 5-year post-diagnostic interval, stratified by their age at the time of surgery. This use case displays

• a more intelligible way of identifying survival in cohorts, and
• a way to compare how survival changes (or rather, it doesn't: the majority of cases, in the 40-49 and 50-59 cohort, have relatively little change in survival, although as time goes on, survival of the 70-79 cohort did experience a significant survival benefit).