Bloodletting

logistic regression
contingency tables
Bloodletting—deliberately withdrawing large amounts of blood from patients—was a common medical practice for centuries, until evidence was finally collected on its efficacy. Use an early observational dataset to explore its use for pneumonia.
Author

Gordon Weinberg

Published

June 8, 2023

Data files
Data year

1828

Motivation

Physician Pierre Charles Alexandre Louis (1787–1872) provided a now-classic analysis which helped to debunk the practice of bloodletting.

In his study, Louis investigated records of pneumonia patients. Bloodletting was so common a medical treatment at the time, every patient Louis investigated had the bleeding procedure performed on them. But some patients had been given the bloodletting later in the course of the pneumonia.

Conventional wisdom at the time was that bloodletting was beneficial, so the patients whose bleeding had been delayed should have fared worse than the patients who had been given bleeding early in the course of the illness.

Data

Each row represents one patient in the study. There were 77 total patients.

Data preview

bloodletting.csv

Variable descriptions

Variable Description
Age The age of the pneumonia patient (only available for patients who died)
DayFirstBled The day of the illness when bloodletting was first performed
FirstBleeding A categorical version of DayFirstBled, defined as “early” if the day of first bleeding as 1 to 4, “late” if day of first bleeding was 5+
NumBleedings The total number of bloodlettings performed on the patient
SickDays The total duration (in days) of the illness
Outcome Whether the patient lived or died from the illness

Questions

  1. One of P.C.A. Louis’s analyses compared the survival rates of patients bled early to those bled late (FirstBleeding). Construct a \(2 \times 2\) contingency table of the outcome (whether the patient lived or died) and whether the patient was bled early or late. Calculate the survival rate for each group. Interpret your results.
  2. Using your contingency table, conducted a \(\chi^2\) test of proportions. Is the survival rate for each group different? State your null hypothesis, test statistic, and p value, and interpret your results.
  3. Use a logistic regression model to predict the outcome based on the day bloodletting was first performed (DayFirstBled). State the coefficient and give a confidence interval. Do patients have better outcomes if they were bled earlier or later?
  4. It’s reasonable to assume that the number of bleedings also matters. Extend your logistic regression to include this (NumBleedings) as a covariate. Interpret both coefficients.

References

Original research: Pierre Charles Alexandre Louis (1836), Researches on the Effects of Bloodletting in Some Inflammatory Diseases, Hilliard, Gray, & Company. Translated by C.G. Putnam.

Data extracted by: Alfredo Morabia (1996). “P.C.A. Lewis and the Birth of Clinical Epidemiology”, Journal of Clinical Epidemiology 49 (12), pp. 1327-1333. https://doi.org/10.1016/s0895-4356(96)00294-6