growth_prediction_shiny_proposal.Rmd

---
title: "R Notebook"
output: html_notebook
---

### "if a hypothetical intervention increased growth by X1 amount at age Y1, it would reduce wasting by X2 amount by age y2"


```{r, include=F}

source(paste0(here::here(), "/0-config.R"))

#Load data
d <- readRDS(paste0(ghapdata_dir, "ki-manuscript-dataset.rds")) %>% filter(whz >= -6 & whz <=6, !is.na(whz), agedays < 731) 

dim(d)

```

  1.  ~30 studies, 61,3695 measures for longitudinal data. Alternatively could use nationally-representative cross-sectional survey data from DHS or MICS (millions of obs)
  2.  -different nutritional and health contexts. Mean growth varies a lot by country
  3.  -ages measured at different intervals and frequencies across studies
  4.  -South Asia overrepresented
  5.  -different covariates in each study


```{r}

p<-ggplot(d, aes(x=agedays)) + geom_histogram()

print(p)

```

```{r}

p<-ggplot(d, aes(x=agedays)) + facet_wrap(~studyid) + geom_histogram()

print(p)

```


```{r}

p<-ggplot(d, aes(x=agedays, y=whz)) + facet_wrap(~studyid) + #geom_point(alpha=0.1) +
  geom_smooth(se=F)

print(p)


```





Shiny app off protected data? Either simple superlearner predictions or a TMLE shift analysis

Weighted SuperLearner?

Considerations in transportability/generalizability?

-Printed red flags based on # of children/studies used in predictions?