Two investigators independently reviewed published studies in MEDLINE, Embase, and Web of Science databases from their inception to October 2018 by using the following index terms: myocardial infarction OR myocardial infarct OR heart attack OR myocardial injury AND cancer incidence OR incident cancer OR new onset cancer OR cancer risk OR carcinogenesis OR tumorigenesis OR oncogenesis. A manual search for additional studies using references of selected retrieved articles was also performed to identify other possible studies. We also conducted searches of conference proceedings from major cardiology and oncology meetings for additional abstracts on the topic. No limitation on language was applied. The last date of the search was October 23, 2018.
Inclusion and exclusion criteria
The inclusion criteria were as follows: 1) observational studies; 2) studies investigating myocardial infarction and cancer incidence; 3) myocardial infarction was clearly diagnosed and new onset cancer occurred after myocardial infarction; and 4) cancer incidence rate or RR, OR, HR or standardized incidence ratio (SIR) with 95% CI were provided.
Exclusion criteria: 1) duplicate publications; 2) publication types were not observational studies; 3) cancer incidence was before or concurrent with MI; and 4) studies were also excluded if relevant data were not reported or extractable.
The quality of each study was assessed by two reviewers. Newcastle Ottawa scale (NOS) generating stars for selection (S), comparability (C) and outcome (O) for cohort and case-control studies were used to evaluate studies’ quality and risk of bias.14 Studies with seven stars or more were assumed to be of high quality, between five and seven stars to be of moderate quality, and less than five stars to be of low quality.
Data extraction was performed independently using a predefined data extract form by two reviewers. The data extracted from each study included the following information: first author’s name, publication year, study design, country, number of patients and controls, gender, cancer incidence, and effect estimates (OR, RR, HR, or SIR) with 95% CI and factors adjusted in the multivariate analysis. If both crude and adjusted values were provided, we only extracted the adjusted ones. Disagreement was resolved by discussion among all investigators.
All analysis was performed using Stata 12.0 software (StataCorp, College Station, TX). The risk of cancer incidence after MI was assessed by ORs and the corresponding 95% CIs. An OR <1 reflects a favorable outcome in MI group compared with controls and indicates lower cancer incidence rate, and vice versa. A two-sided P-value of 0.05 was considered statistically significant for all included analysis. Generic inverse variance (DerSimonian and Laird) method was employed to combine adjusted point estimates and standard errors from each study.15 If effect estimates were presented for a number of categories of exposure, we would combine the corresponding estimates using the method proposed by Hamling et al.16
Q test and I2 statistic were used to ascertain the between-study heterogeneity. A value of I2 of 0%–25%, 25%–50%, 50%–75%, and >75% embodied insignificant, low, moderate, and high heterogeneity, respectively.17 In light of the high likelihood of between-study variance, we used a random-effect model rather than a ﬁxed-effect model. Subgroup analysis would be done based on gender, cancer type, and follow-up time.
Publication bias was assessed by Begg’s test and Egger’s test.18,19 P<0.05 indicated bias, and P>0.05 indicated no publication bias.