Access to Revascularization Among Patients with Acute Myocardial Infarction in New York City—Impact of Hospital Resources

Abstract

Timely revascularization can improve survival in patients with acute myocardial infarction. Identification of factors associated with increased use of revascularization in appropriate patients could improve outcomes. Using New York City hospital discharge records for 1988–1992 and 1998–2002, we determined revascularization rates for patients hospitalized with MI by neighborhood. Odds ratios for revascularization were estimated using a spatial model adjusting for neighborhood sociodemographic characteristics, while accounting for similarities in the rate of revascularization among geographically adjacent neighborhoods. Only 16 out of 112 New York City hospitals performed coronary revascularization. They were located in 14 of 41 neighborhoods. In general, patients living in neighborhoods with higher percentages of patients admitted to hospitals capable of revascularization service were more likely to be revascularized than those in neighborhoods with low percentages of patients admitted to hospitals with revascularization resources. This was true regardless of neighborhood availability of revascularization, after accounting for neighborhood socioeconomic characteristics and patients' clinical status. Revascularization rates in New York City increased from 1988–1992 to 1998–2002 in every neighborhood and as a whole from 103 to 326 per 1,000 hospitalized AMI patients. This increase was not explained by the addition of new revascularization services. Thus, in New York City, where only certain hospitals can perform revascularization, efficient delivery of patients to hospitals with these resources appears to increase the likelihood of revascularization performance among AMI patients without increasing the number of new hospitals capable of revascularization.

Appendix

Specifically, we employed a Bayesian spatial model. Let Y _j denote the number of revascularizations for the jth neighborhood and N _j be the corresponding number of patients hospitalized for AMI; j = 1,...,J. Let θ _j represent the underlying revascularization rate for the jth neighborhood. Given the underlying revascularization rate θ _j , our model assumes the following binomial sampling distribution for the Y _j s:

$${Y_{j} } \mathord{\left/ {\vphantom {{Y_{j} } {\theta j}}} \right. \kern-\nulldelimiterspace} {\theta j} \sim {\text{Binomial}}{\left( {N_{j} ,\theta _{j} } \right)}$$

Considering a logit link function for the revascularization rates θ _j , we assumed that

$${\text{log}}{\left( {\frac{{\theta _{{_{j} }} }}{{1 - \theta _{j} }}} \right)} = \beta _{0} + {\sum\limits_{i = 1}^I {\beta _{{iZi}} + b_{j} } }$$

Where β ₀ represents the logarithm of ‘global’ mean odds of revascularization for New York City as a whole, the β _i s are regression coefficients capturing the effect of neighborhood-level covariates such as average household income or race, and the b _j s are random spatial effects representing log-odds ratio of the corresponding neighborhood after taking into account the effect of neighborhood-level covariates included in the model. Therefore, exp(b _j ) is a relative odds ratio. We put a prior on the b _j s that allows dependency in the likelihood of revascularization among geographically adjacent neighborhoods. We specifically employed the conditional autoregressive (CAR) model. ¹,² This model induces local smoothing by borrowing strength from adjacent neighborhoods. The CAR prior corresponds to the following conditional distribution of b _j :

$${b_{j} } \mathord{\left/ {\vphantom {{b_{j} } {b_{k} }}} \right. \kern-\nulldelimiterspace} {b_{k} } \ne _{j} \sim {\text{Normal}}{\left( {\frac{\lambda }{{1 - \lambda + \lambda n_{j} }}{\sum\limits_{j:k} {b_{k} } },\frac{{\sigma ^{2} }}{{1 - \lambda + \lambda n_{j} }}} \right)},j = 1,...,J$$

To carry out a full Bayesian analysis via the Markov Chain Monte Carlo (MCMC) method, we completed the model specification by assuming the following non-informative priors and hyperpriors:

$$\begin{array}{*{20}l} {{\tau = \sigma ^{{ - 2}} \sim {\text{Gamma}}{\left( {0.001,0.001} \right)}} \hfill} \\ {{\beta _{i} \sim {\text{Normal}}{\left( {0.0,0.001} \right)},i = 0,...,I} \hfill} \\ {{\lambda \sim {\text{Uniform}}{\left( {0,1} \right)}} \hfill} \\ \end{array} $$

Where n _j denotes the number of neighbors for area j, j ∼k indicates that area j is a neighbor of area k and λ is a spatial autocorrelation parameter.