WSO - January 2024 - 52

52
International Journal of Stroke 19(1)
European reference data. Then, we harmonized the effect
sizes for the SNPs on FI and the outcome dataset. For palindromic
SNPs, we used allele frequency information to
resolve strand ambiguity and dropped those with minor
allele frequency close to 0.50. One SNP (rs9275160) associated
with FI was not available in the GISCOME dataset,
and no suitable proxy was found.
Statistical analyses
We performed fixed effect inverse-variance weighted
(IVW) MR as the main analyses unless significant heterogeneity
was present. This approach can provide an accurate
estimate in the absence of horizontal pleiotropy, or when
horizontal pleiotropy is balanced. A number of complementary
MR methods were conducted as complementary analyses,
including the random-effects IVW, weighted median,18
MR-Robust Adjusted Profile Score (MR-RAPS),19 and
MR-Pleiotropy Residual Sum and Outlier (MR-PRESSO)20
approach. The weighted median estimate gives consistent
estimates when at least half of the instrument variables are
valid. The MR-RAPS method was used to provide unbiased
estimates in the presence of many weak instruments.
MR-PRESSO was used to identify potential outliers. These
methods are generally robust to potential violations of
instrumental variable assumptions.
Heterogeneity of each SNP was tested using Cochran's
Q statistic. Leave-one-out plot was created to estimate
whether individual SNPs dominated the overall estimates.
To assess the possibility of reverse causality, we made
Steiger analyses21 to verify the causal direction, as well as
performed reverse MR analyses using genetic instruments
with suggestive significant threshold (p < 5 × 10−6).
Beyond standard analyses, we also carried out additional
sensitivity analyses for further validation and address the
bias in studies of genetic associations with prognosis. First,
we examined the relationship between genetically predicted
fatigue, a characteristic of frailty, and 3-month stroke
outcomes. Genetic variants associated with self-reported
fatigue were ascertained from a UK Biobank GWAS comprising
449,019 individuals.22 Fatigue was assessed based
on the frequency of tiredness/lethargy in the last 2 weeks.
UK Biobank's participants were asked to answer the question,
" Over the last two weeks, how often have you felt
tired or had little energy? " Possible answers were " Not at
all/Several days/More than half the days/Nearly every day/
Do not know/Prefer not to answer. " Participants answering
with " Do not know " or " Prefer not to answer " were
excluded, leading to a four-category variable for fatigue.
Second, a potential challenge to assess prognostic factors
among disease patients is vulnerability to index event bias,
a type of selection bias that happens when studying risk
factors for subsequent events among disease cases. Thus,
we estimated the causal association of FI with ischemic
stroke risk to help to evaluate whether index event bias is
International Journal of Stroke, 19(1)
likely to affect our results of poststroke outcomes. Summary
statistics data for association of SNPs with ischemic stroke
were obtained from MEGASTROKE consortium, including
up to 34,217 ischemic stroke cases and 406,111 strokefree
controls.23 We also performed a replication analysis in
an independent European population using data from the
FinnGen consortium (10,551 ischemic stroke cases and
202,223 controls).24 Finally, we repeated MR analyses
using stroke outcome genetic effects corrected for index
event bias with a correction method, which was described
previously.25
Results were reported for each standard deviation
increment in FI. Post hoc power was calculated at an alpha
level of 0.05 by mRnd (http://cnsgenomics.com/shiny/
mRnd/).26 All statistical analyses were conducted with
TwoSampleMR27 and MR-PRESSO20 packages in R version
4.2.1.
Assessment of pleiotropy
MR-PRESSO global test and MR-Egger intercept were
used to evaluate the evidence of horizontal pleiotropy.
Moreover, we performed multivariable MR (MVMR)
analysis to explore the independent effects of FI on functional
outcome. We considered four covariates associated
with frailty,28 including BMI, C-reactive protein (CRP),
inflammatory bowel disease (IBD), and smoking initiation.
Summary statistics for BMI were derived from the
GIANT (Genetic Investigation of Anthropometric Traits)
consortium.29
The CIWG (Genomic Epidemiology
Inflammation Working Group) was used to obtain genetic
association estimates for CRP.30 Genetic variants for IBD
were from the IIBDGC (The International IBD Genetics
Consortium).31 SNPs for smoking initiation were taken
from the GSCAN (Sequencing Consortium of Alcohol and
Nicotine) consortium.32
Results
Genetic instruments for frailty index
The 14 SNPs were used to instrument FI (Supplementary
Table 5), explaining ≈0.32% phenotypic variation. F statistics
of these instrumental variables ranged from 166 to
212, indicating a low risk of weak instrument bias. We had
100% statistical power in detecting causal effects of FI and
functional outcome after ischemic stroke.
Functional outcome after ischemic stroke
We found MR evidence that genetic liability to frailty in
relation to worse functional outcome after ischemic
stroke in the primary analysis (fixed-effects IVW: odds
ratio (OR) = 5.12; 95% confidence interval (CI) = 1.31-
20.09; p = 0.019; Figure 1). Results from random-effects
http://www.cnsgenomics.com/shiny/
WSO - January 2024

Table of Contents for the Digital Edition of WSO - January 2024
