Incident stroke, risk of dementia

background-image: url(./figs/logo.png)
background-position: 5% 95%
background-size: 35%
class: center, middle

## Effect of incident stroke on the risk of dementia over a period of 10 years of follow-up in a cohort of Asian American and White older adults in California

.pull.right[

L. Paloma Rojas-Saunero MD, PhD
<br>
Postdoctoral scholar
<br>
**Mayeda Research Group, Department of Epidemiology**

]

---

## Motivation

- Several studies have shown that poststroke cognitive impairment (PSCI) is frequent, and a large percentage of stroke survivors do no fully recover from PSCI

- Although there are several modifiable risk factors that can reduce both the risk of stroke and dementia, there is a need to understand in more depth the role of the acute vascular event in the predisposition for Alzheimer’s disease and related dementias.

- Previous studies looking at the association between stroke - dementia had major limitations (used prevalent stroke, incident stroke as time-fixed, limited information on death as competing event

- Very limited information on the effect for asian-american population

---
## Research question

- What is the direct effect of incident stroke in the risk of dementia across different Asian American ethnicities and White populations, if we could have prevented death? (*Controlled direct effect*)

- What is the total effect of incident stroke in the risk of dementia  across different Asian American ethnicities and White populations? (*Total Effect*)
---
## Study population

- **Study sample:**

+ KPNC members who participated on the California Men’s Health Study (CMHS) or the Kaiser Permanente Research Program on Genes, Environment and Health Survey (RPGEH) who self-identified as Asian Americans or white.

- **Eligibility criteria:**

+ With no history of stroke at baseline (survey year)
 + With no history of dementia at baseline
 + From 60 to 89 years old at baseline
 + With information on ethnicity
 + With follow-up

---
## Study Design

- **Exposure:** Incident stroke (ischemic stroke, hemorrhagic stroke)

- **Outcome:** Incident dementia diagnosis (Alzheimer’s disease, vascular dementia, and non-specific dementia diagnosis)

- **Time zero/Baseline:** Time of survey

- **End of follow-up:** Time of dementia diagnosis, time of death prior to dementia diagnosis (competing event), time end of membership (censoring event) or turning 90 years old.

- **Time-scale:** Years of follow-up

---
## Covariates

.pull-left[

**Time-fixed covariates**

- Nativity status
- Educational attainment
- Health status
- Age at survey
- Sex/gender
- Smoking status

]

.pull-right[

**Time-varying covariates**

- Systolic blood pressure (median value/year)
- BMI (median value/year)
- Lipids
- Incident diabetes
- Incident hypertension
- Incident myocardial infarction
- Incident congestive heart failure
- Incident cancer

]

---
## Statistical Analysis

**Controlled direct effect**
- IPTW for the probability of developing stroke at each time-point based on time-fixed and time-varying covariates.

- IPCW for the probability of end of membership over follow-up (IPCWm).

- IPCW for death over follow-up (IPCWd). This is to satisfy the independent censoring assumption between dementia and death. (_Controlled direct effect_)

- Plug in IPTW x IPCWm x IPCWd in the Kaplan-Meier estimator

- Calculate cumulative incidence of dementia, Risk ratio and Risk difference at 10 years. Bootstrap confidence intervals.

---

## Statistical Analysis

**Total effect**

- Plug in IPTW x IPCWm in the Aalen-Johannsen estimator

- Calculate cumulative incidence of dementia, Risk ratio and Risk difference at 10 years. Bootstrap confidence intervals.

---
## Example

.center[
<img src=./figs/cif_curves_filip.png width="100%"/>
]

---
## Covariates used for the models

- IPTW:
  + FU yr (bspline), baseline age (bspline)
  + Baseline covariates: gender, nativity, education, general health, smoking status, prevalent acute myocardial infarction
  + Time-varying covariates: diabetes, hypertension, incidence acute myocardial infarction, congestive heart failure, peripheral vascular disease, BMI, SBP

- IPCWd: Death weights. 
  + Same variables as IPTW and t-v stroke indicator

- IPCWm: End of membership weights
  + Survey age, education, and general health

---
## Results

.center[
<img src=./figs/stroke_counts.png width="80%"/>
]

---
### Direct effect of stroke on the risk of dementia

.center[
<img src=./figs/rr_de.png width="80%"/>
]

---
### Direct effect of stroke on the risk of dementia

.center[
<img src=./figs/rd_de.png width="80%"/>
]

---
### Total effect of stroke on the risk of dementia

.center[
<img src=./figs/rr_te.png width="80%"/>
]

---
### Total effect of stroke on the risk of dementia

.center[
<img src=./figs/rd_te.png width="80%"/>
]

---

### Weights assessment - IPTW for Chinese population

.center[
<img src=./figs/iptw_chi.png width="95%"/>
]

---
### Weights assessment - IPCW for Chinese population

.center[
<img src=./figs/ipcw_chi.png width="95%"/>
]

---
## Take-aways

- The incidence of stroke is large across all Asian-American ethnicities and in the White population.

- There is a large effect of stroke on the risk of dementia if we treat death as a censoring event (_as if we could have prevented it_).

- There are several factors that could explain the heterogeneity of the effects across subgroups.

---
## Discussion points

- Because we use use discrete time in the analysis, this results in rounding and we loose events. Suggestions for sensitivity analysis?

- How can we prevent the potential reverse causation of those who have a dementia diagnosis very close to the stroke event?

--
 
- Smaller time-frame to look at stroke? What would be a reasonable time-frame with the trade off of loosing outcomes?

---
## Funding

This work was supported by NIA R01AG063969, "Alzheimer's disease and related dementias in a diverse cohort of Asian Americans".

---
class: center, middle

### Thank you, Gracias!

---

class: center, middle

![](https://media.giphy.com/media/XHVmD4RyXgSjd8aUMb/giphy.gif)