Using R to explore outpatient trajectories from linked healthcare data

class: center, middle, inverse, title-slide

# Using R to explore outpatient trajectories from linked healthcare data
### Fiona Grimm, The Health Foundation NHS-R conference 2019 @fiona_grimm @HealthFdn
### 2019/11/04

---

# Outpatient care redesign in the NHS Long Term Plan

.pull-left[

<img src="FionaGrimm_NHSR_files/figure-html/unnamed-chunk-4-1.png" width="550px" style="display: block; margin: auto;" />
]

--
 
.pull-right[
 
"...over the next five years patients will be able to avoid up to a third of face-to-face outpatient visits..."

### Strategies to achieve this

* Avoid referrals by supporting GPs

* Replace face-to-face with virtual appointments
* Fewer and/or more flexible follow-ups

What will be the effect on demand and how long will it take?
]

---
class: inverse, middle, center

# How can operational research techniques inform policy?

---
## Proof of concept: testing alternative models of diabetes outpatient care

.large[
### The approach
1. Primary care data (CPRD) linked to HES for a cohort of diabetes patients

2. Characterise follow-up frequencies

3. Create transition probability matrix
]

---

.pull-left[

### From patient-level follow up patterns..

]
---

.pull-left[

### From patient-level follow up patterns..

<img src="FionaGrimm_NHSR_files/figure-html/unnamed-chunk-6-1.png" width="450px" style="display: block; margin: auto;" />
]

--
.pull-right[
### ..to time interval between follow-ups

<img src="FionaGrimm_NHSR_files/figure-html/unnamed-chunk-7-1.png" width="450px" style="display: block; margin: auto;" />
]

---

.pull-left[

### From patient-level follow up patterns..

<img src="FionaGrimm_NHSR_files/figure-html/unnamed-chunk-8-1.png" width="450px" style="display: block; margin: auto;" />
]

.pull-right[
### ..to time interval between follow-ups

<img src="FionaGrimm_NHSR_files/figure-html/unnamed-chunk-9-1.png" width="450px" style="display: block; margin: auto;" />
 
]

---

.pull-left[

### From patient-level follow up patterns..

<img src="FionaGrimm_NHSR_files/figure-html/unnamed-chunk-10-1.png" width="450px" style="display: block; margin: auto;" />
]

.pull-right[
### ..to time interval between follow-ups

<img src="FionaGrimm_NHSR_files/figure-html/unnamed-chunk-11-1.png" width="450px" style="display: block; margin: auto;" />
 
]

--
.pull-right[
 map to follow-up categories 
 create transition probability matrix

]

---
## Proof of concept: the impact of policy scenarios on diabetes outpatient care

.large[

### The approach
1. Primary care data (CPRD) linked to HES for a cohort of diabetes patients

2. Characterise follow-up frequencies

3. Create transition probability matrix

4. Model the overall volume of appointments    
using stochastic analytical models (Markov chains) and/or simulations

5. Test the effect of different scenarios  ]

---
class: inverse, center, bottom
background-image: url(https://creativepool.com/files/candidate/portfolio/full/1118419.jpg)
background-size: 40%

# some great #rstats tools that helped with this analysis

---
# {tidylog} expects nothing and gives great feedback

```r
library(tidyverse)

summary <- mtcars %>% 
 mutate(mpg_by_wt = mpg / wt) %>% 
 filter(cyl > 4) %>% 
 select(mpg_by_wt, cyl)
```

---
# {tidylog} expects nothing and gives great feedback

```r
library(tidyverse)
*library(tidylog)

summary <- mtcars %>% 
 mutate(mpg_by_wt = mpg / wt) %>% 
 filter(cyl > 4) %>% 
 select(mpg_by_wt, cyl)
```

```
## mutate: new variable 'mpg_by_wt' with 32 unique values and 0% NA
```

```
## filter: removed 11 rows (34%), 21 rows remaining
```

```
## select: dropped 10 variables (mpg, disp, hp, drat, wt, …)
```
--

.pull-left[.content-box-green[
Zero effort!   
Saves time and helps avoid mistakes 
]]

.pull-right[.content-box-red[
Beware when iterating:   
printing console output might be slow]]

---
# Tools for filtering

```r
library(tidyverse)
library(lubridate)
library(nycflights13)
```

.pull-left[
### By time period

```r
start <- ymd('2013-01-01')
end <- start + days(2)

my_interval <- interval(start, end)

flights %>% 
* filter(time_hour %within% my_interval)
```
]

.pull-right[
### By another table

```r
top5_carries <- flights %>% 
 group_by(carrier) %>% 
 count() %>% 
 arrange(desc(n)) %>% 
 head()

flights %>% 
* semi_join(top5_carries, by = 'carrier')
```
]
---

# Tools for summarising

.pull-left[
### Cross-tabs with {janitor}

```r
library(janitor)

mtcars %>% 
* tabyl(gear, cyl) %>%
  adorn_percentages() %>% 
  adorn_pct_formatting() %>% 
  adorn_rounding() %>% 
  adorn_title()
```

```
##         cyl            
##  gear     4     6     8
##     3  6.7% 13.3% 80.0%
##     4 66.7% 33.3%  0.0%
##     5 40.0% 20.0% 40.0%
```
]

.pull-right[
### Baseline characteristics with {tableone}

```r
library(tableone)

CreateTableOne(data = iris, 
               vars = c('Sepal.Length', 
                        'Sepal.Width', 
                        'Species'))
```

```
##                           
##                            Overall     
##   n                         150        
##   Sepal.Length (mean (SD)) 5.84 (0.83) 
##   Sepal.Width (mean (SD))  3.06 (0.44) 
##   Species (%)                          
##      setosa                  50 (33.3) 
##      versicolor              50 (33.3) 
##      virginica               50 (33.3)
```
]
---

.pull-left[
# Acknowledgements

### Project team    
* .bold[Meetali Kakad], University of Oslo   
* Martin Utley, UCL Clinical Operations Research Unit   
* Sarah Deeny, Health Foundation

### Funding
* The Health Foundation
* Research Council of Norway

### Clinical input   
* Sarita Naik, Consultant Diabetologist, UCLH    
* Ali Chakera, Consultant Diabetologist, Brighton and Sussex University Hospital   
]

.pull-right[
# Get in touch
.red[@fiona_grimm]

.red[@tali_md]

# Find our R code on GitHub
.red[hfanalyticslab.github.io]
![](FionaGrimm_NHSR_files/figure-html/unnamed-chunk-21-1.png)

]