All 4 books by Edward Tufte now in
paperback editions, $100 for all 4
Visual Display of Quantitative Information
Beautiful EvidencePaper/printing = original clothbound books.
Only available through ET's Graphics Press:
catalog + shopping cart
All 4 clothbound books, autographed by the author $150
catalog + shopping cart
Edward Tufte e-books
Immediate download to any computer:
Visual and Statistical Thinking $2
The Cognitive Style of Powerpoint $2
Seeing Around + Feynman Diagrams $2
Data Analysis for Politics and Policy $2catalog + shopping cart
Edward Tufte one-day course,
Presenting Data and Information
San Francisco CA, December 3, 4, 5
San Jose CA, December 7
In January 2000, Dr Harold Shipman, a former family doctor in Manchester, England was found guilty of murdering 15 of his patients. It is suspected that he killed more, possibly as many as 250.
Today he was found dead in prison.
Following his trial, the Chief Medical Officer for England commissioned an audit of Shipman's clinical practice for the years from his appointment in 1974 to his arrest in 1998. That detailed review shows how the deaths of Shipman's patients displayed a number of unusual characteristics with respect to time and place. Perhaps one of the most chilling was the early afternoon peak (during home visits) when the proportion of deaths for Shipman (14% at 2pm) was about 7 times higher than a comparison group of doctors.
To add to the tragedy for victims, their families and other professionals was the fact that this information was already 'in the system' all along but, of course, was not recorded, aggregated, tabulated or graphed in a way that might have compelled investigation.
The difference between life and death sometimes lies (continuously hidden) in the data.
The graph is Figure 5.2 on page 40 of the report.
-- Mark Reilly (email)
"So much data ... so little information"! An excellent example, Mr. Reilly.
Even the tabular data in table 2.1 (page 19) displayed on a simple control chart would have pointed to Shipman being a "special cause" deserving critical attention soon after Shipman began his practice.
A tragic analytic conclusion to this story: the closing statement by Chief Medical Officer Professor Liam Donaldson (page 4): "Everything points to the fact that a doctor with the sinister and macabre motivation of Harold Shipman is a once in a lifetime occurrence."
HOW DO WE KNOW THIS? Since this episode, has CMO Donaldson implemented processes to catch such doctors early in their practice careers? What about the credentialing agencies that must renew licenses? What efforts have they taken to ensure the integrity of the profession? What (if anything) has been done to change the system?
-- Michael Round (email)
Mark Reilly has linked us to an intriguing and ghastly report.
But the CMO document muddles along slowly, working through all those marginal distributions at the beginning before getting on with the real detective work. For comparison, it would be instructive to see a tighter presentation of the evidence done in the style of the Morbidity and Mortality Weekly Report, the often-superb publication of the Centers for Disease Control and Prevention. The MMWR routinely describes evidence about patterns of death attributable to a particular cause, like the CMO document.
Another approach would be an article in a medical journal, such as The Lancet. The CMO report will soon become a fugitive publication, hard to find, too long to print out; better to get it into a tighter, more accessible, and more scientific form--such as the MMWR or The Lancet. In such publications, the report would have to meet higher standards of evidence and argument than in a government publication.
The consequence of the MMWR/Lancet approach might be to shift public discussion away from the tabloid approach (analysis of the personality of the evil doctor and of mug shots that invite trying to detect evil by facial characteristics). Instead, the long-term issues of this case are: What kinds of statistical evidence indicate problems in medical practice? Are there early warning signals? How can the performance of complex systems be monitored? This in turn would help in identifying and perhaps preventing future episodes of this type, as well as providing evidence that would detect other types of poor performance.
The loose CMO report may tend to turn journalists loose; a tight scientific report in a major medical publication might tighten up the journalism, covering the story not only a crime story but also as a public policy story.
-- Edward Tufte
A couple of things strike me and as a forensic scientist, I thought I'd chip in.
First, Shipman is certainly not a "once in a lifetime occurence." Nurses and other medical assistants are caught every so often killing off patients in a pseudo-Munchausen-by-proxy series of killings. Being on call when the death "occurs," they gain the thrill of the heroic attempt to save the patient's life and the sympathy at "having lost another one." Really just one step closer than being Burke and Hare (Irish immigrants turned mass murderers making money by providing bodies to a Dr. Knox. Burke developed a method of suffocating people that left little or no sign of violence on them--he would hold their nose and mouth closed while they slept, suffocating them. In 1828, Burke and Hare murdered 16 people by suffocation and sold them all to Dr. Knox for dissection for £7 to £10.)
Second, it's nearly impossible to predict the actions of one person like Shipman. Forensic science is a historical science--we can only reconstruct past criminal events through the evidence collected at the scene--and therefore makes predictions quite difficult. Witness the variations offered by forensic "profilers" about the DC sniper (couldn't have been more wrong). Predicting the next Shipman is almost impossible. What might work better is some way to find a heuristic that is key to deciding if further investigation is warranted. These kind of heuristics are being studied at the Max Planck Institute for Human Development in Berlin by Gigerenzer, Todd, and others:
Finally, graphics as applied to crime have been useful if ugly (http://www.ojp.usdoj.gov/nij/maps/images/bb4.gif). "Crime mapping" helps agencies plan, pinpoint, and allocate resources but, as you can see from the example above, they need lots of help in displaying this information. The ellipses in this map were drawn by hand, if you can believe it, to emphasize the clusters. Can't see the crime for the streets, hardly. [links updated March 2005]
-- Max Houck (email)