Table 4.

Study-Specific Factors That Influence Confounding Relative to Other Studies

Study	Reasons Confounding May Be Less	Reasons Confounding May Be Greater
Chen et al5	Good measures of comorbidities and functional status. Eliminated patients with terminal illness. Compared board certified physicians.
Francis et al7	Instrumental variable related to hospital distance was used to reduce confounding.	Strong confounding of physician type with patient age.
Norcini et al18	Excellent assessment of comorbidities.	No information on functional status or race. Strong confounding of physician type with patient age.
Frances et al23	Good measures of comorbidities and functional status.	Generalists patients have very high risk. Patient location not assessed.
Nash et al9	Analyzed patients in two age strata. Adjusted for some demographic variables using propensity score.	No information on functional status or race.
Casale et al10	Excellent assessment of comorbidities.	No information on functional status or race. Strong confounding of physician type with patient age.
Ayanian et al6	Excluded patients 80 or older.	No information on functional status or race.
Nash et al13		No information on functional status or race.
Jollis et al22		No comorbidity or functional status.