Table 3.

Summary of Studies on the Effect of CON on Mortality for Coronary Artery Bypass Graft (CABG) Procedures

STUDY	DESIGN	STATES	YEARS	POP-ULATION	CONTROLS	ESTIMATED EFFECT OF CON ON MORTALITY						NOTES
						Mortality measure	Mean change1	Stand-ard error	Net change, deaths per 1000 patients2	Pop-ulation weight3	Recency weight4
Total Population Studies
Ho [65]	Retro-spective cohort	18 CON vs. 8 no CON	1988–2000	All	State fixed effects, patient characteristics	Inpatient mortality	−2.6%	0.8%	−1.00	77.1%	50.0%	Only one state in sample dropped CON during the study period. Finds no mortality effect of CON on PTCA. SE imputed from reported t statistic.
DiSesa et al. [52]	Retro-spective cohort	27 CON vs. 24 no CON	2000–2003	All	State and hospital fixed effects, patient controls	Operative mortality	−4.9%	5.7%	−1.25	100.0%	100.0%
Robinson et al. [42]	Pre-post	PA	1994–1999	All	Patient characteristics	Inpatient mortality	0.0%	0.9%	0.00	0.0%	50.0%	After CON lifted, actual mortality matched expected mortality for both old and new cardiac programs; uses same PHC4 data as Kolstad
Kolstad [39]	Pre-post	PA	1994–2003	All	Compares incumbent hospitals to new entrants		2.9%	0.9%	0.62	4.3%	100.0%	Kolstad calculates that 11 deaths are averted annually by CON repeal. His Table 1.1 shows that in 2000–2003, the 40 incumbent hospitals performed an average of 349 CABGs, for a total of 13,960 (RAMR = 2.17%) while 24 new entrants performed an average of 160 (RAMR = 2.04%). Weighted average mortality = 2.14% vs. 2.20% if 11 additional deaths had occurred.
Cutler et al. [53]	Pre-post	PA	1994–2003	All	Compares incumbent hospitals to new entrants		2.9%	0.9%	0.62	0.0%	100.0%	Essentially the same paper as Kolstad [39]
Weighted average:									−1.13
Medicare Patient Studies
Vaughan-Sarrazin et al. [57]	Retro-spective cohort	27 continuous CON vs. 18 no CON	1994–1999	Medicare (excludes managed care)	Patient characteristics	In-hospital mortality	−17.3%	2.6%	− 8.70	77.1%	25.0%	States without CON exhibited CABG higher mortality (OR = 1.22) than states with continuous CON; this implies CON is associated with an 17.3% decrease in mortality rates, derived algebraically. No effect in intermittent CON states
Popescu et al. [58]	Retro-spective cohort	27 CON vs. 23 no CON	1998–2000	Medicare	Patient characteristics	30-day all-cause mortality	−5.0%	1.0%	−8.90	100.0%	50.0%
DiSesa et al. [52]	Retro-spective cohort	27 CON vs. 24 no CON	2001	Medicare patients age 65 and older (excludes managed care)	State and hospital fixed effects, patient controls	Operative mortality	−0.3%	4.9%	−0.10	100.0%	75.0%
Popescu, Vaughan-Sarazin and Rosenthal [55]	Retro-spective cohort	27 CON vs. 24 no CON	2000–2003	Medicare (age 68+)	Patient characteristics	30-day all-cause mortality	0.0%	1.5%	0.00	100.0%	100.0%	Vaughan-Sarazin is co-author on this paper; her most recent work, using the most recent data she uses, finds zero effect (a true 0.00 estimate; not just statistically insignificant)
Ho et al. [59]	Retro-spective cohort	27 continuous CON vs. 7 dropped CON	1989–2002	Medicare patients age 65 and older (excludes managed care)	State fixed effects, extensive hospital and patient controls	Procedural mortality	10.8%	3.3%	5.20	63.9%	100.0%	Dropping CON reduces mortality at first, but the effect dissipates 5 years after CON is removed
Weighted average:									−0.93
Popescu, Vaughan-Sarazin and Rosenthal [55]	Retro-spective cohort	27 CON vs. 24 no CON	2000–2003	Medicare (age 68+)	Patient characteristics	30-day all-cause mortality	−4.2%	2.6%	−7.31	100.0%	100.0%	States with stringent CON lower mortality but effect is of borderline statistical significance

¹Mean change in probability of death, calculated by authors using data reported at original source

²All figures calculated by authors: 1000 x (Mean Mortality Rate in CON States) x (1–1/(1 + Mean Change)) using data on the mean mortality rate for the relevant population and mortality measure shown as reported at original source

³Population weights represent the fraction of the theoretical population of interest included in a study. All figures are calculated by authors based on the total number of CABG surgeries in 2008, allocated to states based on 2008 Census figures on total adult population age 18 and older (for total population studies) and total population age 65 and older (for Medicare patient studies). A weight of zero has been assigned to studies that either duplicate other reported findings or have been entirely superseded by analyses using the same (overlapping) data source but with additional newer years of data

⁴Recency weights are calculated to provide greater weight to results that rely on more recent data and/or improved methods