analysis.Rmd

---
title: 'Analysis and visualisations for "Smartphone apps for the treatment of mental disorders: a systematic review" (preprint)'
author: "Carlos Granell, Juana Breton-Lopez, Sven Casteleyn, Diana Castilla, Laura Diaz, Adriana Mira, Ignacio Miralles, William Van Woensel"
date: "03 June, 2019"
output:
  html_document:
    df_print: paged
    toc: yes
urlcolor: blue    
---


```{r load_libraries, echo=FALSE, message=FALSE, warning=FALSE}
library(here)
library(tidyverse)
library(cowplot)    # publication-ready theme for ggplot2
library(kableExtra)
library(forcats)
library(purrr)
library(scales)
library(knitr)
```

## About

This computational essay is versioned in a public [git](https://git-scm.com/) repo: [https://github.com/cgranell/apps-treatment-mental-disorders](https://github.com/cgranell/apps-treatment-mental-disorders). This R Markdown file, `analysis.Rmd`, contains the code to produce the final figures and tables of the following [paper in JMIR Preprints](https://preprints.jmir.org/preprint/14897). 

> Miralles I, Granell C, Díaz-Sanahuja L, Van Woensel W, Bretón-López J, Mira A, Castilla D, Casteleyn S
> Smartphone apps for the treatment of mental disorders: a systematic review.
> JMIR Preprints. 03/06/2019:14897
> DOI: 10.2196/preprints.14897

\newpage

## Data 

```{r all_data, echo=FALSE}
load(file = here("data", "all_data.rda"))

#' Replace existing factors levels
levels(all_data$val_asstype)[levels(all_data$val_asstype)=="EFFECT AND USABILITY/USER EXPERIENCE"] <- "EFFECT AND USABILITY/UX"
levels(all_data$val_asstype)[levels(all_data$val_asstype)=="USABILITY/USER EXPERIENCE"] <- "USABILITY/UX"

n_md <- nlevels(all_data$md)
n_papers = nrow(all_data)

n_assessment_no <- nrow(filter(all_data, val_asstype == "NO ASSESSMENT"))
n_assessment_yes <- n_papers - n_assessment_no

n_assessment_effect <- nrow(filter(all_data, val_asstype=="EFFECT"))
n_assessment_effectux <-nrow(filter(all_data, val_asstype=="EFFECT AND USABILITY/UX"))
n_assessment_ux <-nrow(filter(all_data, val_asstype=="USABILITY/UX"))

# Keep the same order of mental disorder for all charts 
temp_for_order <- 
    all_data %>% 
    group_by(md_id, md_desc) %>%
    summarise(number_cases = n()) %>%
    mutate(proportion = number_cases/n_papers) %>% 
    arrange(desc(proportion), md_id) 

# convert to factor to retain sorted order of mental disorders 
temp_for_order$md_desc <- factor(temp_for_order$md_desc, levels=unique(temp_for_order$md_desc))  

# Save ordered mental disorders
md_all_ordered <- levels(temp_for_order$md_desc)

default_palette <- c("NO ASSESSMENT"="#AF8DC3", "USABILITY/UX"="#D9F0D3", "EFFECT AND USABILITY/UX"="#7FBF7B", "EFFECT"="#1B7837")
```

The final number of surveyed papers is `r n_papers`. For each paper, we have extracted `r ncol(all_data)` items. Each item is described in the methods section of the [paper](https://preprints.jmir.org/preprint/14897). About `r percent(n_assessment_yes/n_papers)` (N=`r n_assessment_yes`) of papers reported some kind of assessment, whereas `r percent(n_assessment_no/n_papers)` (N=`r n_assessment_no`) reported no assessment at all. Looking closer at the type of assessment: `r percent(n_assessment_effect/n_papers)` (N=`r n_assessment_effect`) focus on the effect of intervention on clinical symptomology; `r percent(n_assessment_effectux/n_papers)` (N=`r n_assessment_effectux`) report a mix of effect and usability / UX assessment; `r percent(n_assessment_ux/n_papers)` (N=`r n_assessment_ux`) focus solely on usability / UX.

\newpage

## Figures and tables

### Figure 2: Temporal trend of amount of articles published and assessment type. The grey bar in the top figure denotes the estimated number of papers in 2018 (three months covered).

It combines two plots. The top one is a _stacked bar chart_ over years, adn the vaue of each group (assessment type) is in absolute counts. The bottom one is a _proportional stacked area chart_ in which the sum of each year is always equal to hundred, and the value of each group (assessment type) is in percentages.

```{r fig2_tempdist, echo=FALSE, dpi=600, fig.width=10, fig.height=22, fig.asp=1}
cols <- c("md", "md_id", "md_desc", "val_asstype", "year")

data_plot_tempdist <- 
    all_data %>%
    select(cols)

data_plot_tempdist$val_asstype <- forcats::fct_relevel(data_plot_tempdist$val_asstype, 
                                             c("NO ASSESSMENT", "USABILITY/UX", "EFFECT AND USABILITY/UX", "EFFECT"))

data_plot_tempdist <- 
    data_plot_tempdist %>%
    group_by(year, val_asstype) %>%  # first create counts for each group
    summarise(number_cases = n()) %>%
    mutate(total_cases_per_year = sum(number_cases),
           proportion = number_cases/total_cases_per_year,
           proportion_lbl = paste0(round(proportion*100,1), "%"))

# Not run
# display.brewer.all()
# display.brewer.pal(n = 6, name = "PRGn")
# brewer.pal(n = 6, name = "PRGn")[2:5]
pal <- default_palette

# an estimate of published papers to this matter for 2018 
prediction_2018 <- 52

plot_tempdist_stakedbarchart <- 
    data_plot_tempdist %>%
        ggplot(aes(x=year, y=number_cases, fill=val_asstype)) + 
        geom_bar(stat="identity") +
        geom_text(aes(label=number_cases), size=2.5,position=position_stack(vjust = 0.5)) +
        labs(#title="Count of papers per year,, colored by assessment type", 
             x="", 
             y="# of papers") +
        scale_fill_manual(values = pal) +
        scale_x_continuous(breaks = seq(2013,2018, by=1)) +
        guides(fill="none") +
        # Prediction 2018
        annotate("rect", alpha = 0.2, 
            xmin = 2018 - 0.45, xmax = 2018 + 0.45,
            ymin = 13, ymax = prediction_2018, linetype="dashed") +
        theme_minimal() +
        # Change the line type and color of axis lines
        theme(axis.line = element_line(colour = "darkgray", size = 0.5, linetype = "solid")) +
        theme(panel.grid.minor = element_blank()) +
        theme(panel.background = element_blank()) +
        theme(plot.margin=unit(rep(20, 4), "pt"))

# plot_tempdist_stakedbarchart

plot_tempdist_areachart <- 
    data_plot_tempdist %>%
        ggplot(aes(x=year, y=proportion, fill=val_asstype)) + 
        geom_area() +
        labs(#title="Temporal distribution of papers per assessment type",
        x="Year", 
        y="Percentage of papers [%]", 
        caption="Source: authors") + 
        guides(fill=guide_legend(title="Assessment Type", nrow=2)) + # modify legend title
        scale_fill_manual(values = pal) +
        scale_y_continuous(expand=c(0,0), labels=scales::percent_format(), breaks=seq(0, 1.00, by=0.20), limits=c(0, 1.00)) +
        theme_minimal() +
        theme(legend.title = element_text(size=11), 
              legend.position = "bottom",
              legend.background = element_rect(color = "darkgray", size = 0.5, linetype ="solid"),
              legend.key = element_blank()) +
        # Change the line type and color of axis lines
        theme(axis.line = element_line(colour = "darkgray", size = 0.5, linetype = "solid")) +
        theme(panel.grid.major = element_blank()) + 
        theme(panel.grid.minor = element_blank()) +
        theme(panel.background = element_blank()) +
        theme(plot.margin=unit(rep(20, 4), "pt"))

#plot_tempdist_areachart

# Put labels = C("A", "B") to identify subfigures 
plot_tempdist <- plot_grid(plot_tempdist_stakedbarchart, plot_tempdist_areachart, labels = NULL, ncol= 1, align = "v") 
plot_tempdist

plot_tempdist_file_name <- "fig2_tempdist.png"
cowplot::ggsave2(plot = plot_tempdist, filename = plot_tempdist_file_name, device = "png", 
                 path = here::here("figs"), scale = 1, 
                 width = 20, height = 28, units = "cm", dpi = 600)

```

\newpage

### Figure 3: Distribution of articles per mental disorder, categorized according to assessment type. The embedded pie chart shows the proportional distribution of types of assessment over all mental disorders.

The main plot is a _stacked barchart_ colored by assessment type along with a _lollipop chart_ (white circle at the end of each bar) that shows the total percentatge of each mental disorder. Note that we draw two main colors in the legend: purple represents no assessment at all, the range of greens denotes distinct types of assessment. A _pie chart_ is embedded to show the proportional distribution of types of assessment over all mental disorders.

```{r fig3_barchart, echo=FALSE,dpi=600,fig.width=10,fig.asp=0.65}
cols <- c("md", "md_id", "md_desc", "val_asstype")

data_plot_barchart <- 
    all_data %>%
    select(cols)

data_plot_barchart <- 
    data_plot_barchart %>% 
        group_by(md_id, md_desc, val_asstype) %>%  # first create counts for each group
        summarise(number_cases = n()) %>%
        mutate(proportion = number_cases/n_papers) %>% 
        ungroup() %>%
        group_by(md_id) %>%
        mutate(total_cases = sum(number_cases),
               proportion_sum = sum(proportion),
               proportion_lbl = paste0(round(proportion_sum*100,1), "%")) %>% 
        ungroup() %>%
        mutate(lbl = paste0(round(number_cases/total_cases*100,1), "%")) %>% 
        arrange(desc(proportion_sum), md_id)

# convert to factor to retain sorted order in plot.
data_plot_barchart$md_desc <- factor(data_plot_barchart$md_desc, levels=unique(data_plot_barchart$md_desc))  

data_plot_barchart$val_asstype <- forcats::fct_relevel(data_plot_barchart$val_asstype, 
                                            c("NO ASSESSMENT", "USABILITY/UX", "EFFECT AND USABILITY/UX", "EFFECT"))
                                            
pal <- default_palette

lbls <- distinct(data_plot_barchart, md_desc, proportion_sum, proportion_lbl)

top_proportion <- sum(lbls[1:6, "proportion_sum"]) 
top_lbl <- paste0(round(top_proportion*100,1), "%")

plot_barchart <- 
    data_plot_barchart %>%
      ggplot(aes(x=md_desc, y=proportion, fill=val_asstype)) + 
        geom_bar(stat="identity") +
        labs(#title="Distribution per mental disorder and assessment type", 
             #subtitle = "Total percentatge per mental disorder within each point",
             x="Mental disorders", 
             y="# of papers and total percentage", 
             caption="Source: authors") + 
        geom_point(aes(y=proportion_sum), size=6, color="white", show.legend = F) +  
        geom_text(aes(label=number_cases), size=2.5, position=position_stack(vjust = 0.4)) +
        # Percentatge inside point
        annotate("text", x = lbls$md_desc, y = lbls$proportion_sum,
                 label = lbls$proportion_lbl, color = "black", size=2, hjust = 0.4, vjust = 0.2) +
        # Arrow to indidate  Top6 mental disorders 
        annotate("text", x = "Schizophrenia spectrum and\n other psychotic disorders", y = .16,
                 label = top_lbl, color = "black", size = 3, hjust = -0.1, vjust = 1.2) +
        geom_segment(aes(x = "Schizophrenia spectrum and\n other psychotic disorders", 
                         xend = "Substance-related and\n addictive disorders", 
                         y = .18, 
                         yend = .18),
                         arrow = arrow(length = unit(0.5,"cm")), color = "black") +
        geom_segment(aes(x="Schizophrenia spectrum and\n other psychotic disorders", 
                         y=0.11, 
                         xend="Schizophrenia spectrum and\n other psychotic disorders", 
                         yend=0.18), color="black") +
        coord_flip() +
        scale_fill_manual(name="Assessment Type", values = pal) + 
        scale_y_continuous(expand=c(0,0), labels=scales::percent_format(accuracy=1), breaks=seq(0, 0.21, by=0.02), limits=c(0, 0.21)) +
        # Which legend to show
        guides(fill=guide_legend(title="Assessment Type", nrow=2)) + # modify legend title
        theme_minimal()  +
        theme(legend.title = element_text(size=11),
              legend.position = "bottom",
              legend.background = element_rect(color = "darkgray", size = 0.5, linetype ="solid"),
              legend.key = element_blank()) +
        # Change the line type and color of axis lines
        theme(axis.line = element_line(colour = "darkgray", size = 0.5, linetype = "solid")) +
        theme(panel.grid.major = element_blank()) + 
        theme(panel.grid.minor = element_blank()) +
        theme(panel.background = element_blank()) +
        theme(plot.margin=unit(rep(20, 4), "pt"))

# plot_barchart


data_plot_piechart <-
    data_plot_barchart %>%
    group_by(val_asstype) %>%
    summarise(total_asstype = sum(number_cases)) %>%
    mutate(lbl = paste0(round(total_asstype/n_papers*100,1), "%"),
           cumulative= cumsum(total_asstype),
           midpoint = cumulative-total_asstype/2) %>% 
    arrange(desc(total_asstype))

data_plot_piechart$val_asstype <- forcats::fct_relevel(data_plot_piechart$val_asstype, 
                                               c("EFFECT", "USABILITY/UX", "NO ASSESSMENT", "EFFECT AND USABILITY/UX"))
plot_piechart <- 
    data_plot_piechart %>%
    ggplot(aes(x="", y=total_asstype, fill=val_asstype)) +
    geom_bar(width=1, stat = "identity") +
    coord_polar(theta="y", start=0, direction = 1) +
    geom_text(aes(y=midpoint, label=lbl), size=3, show.legend = F) +
    scale_fill_manual(name="Assessment Type", values = pal) + 
    # Which legend to show
    guides(fill="none") +
    theme_minimal() +
    theme(legend.title = element_text(size=9)) +
    theme(axis.text = element_blank(),
          axis.ticks = element_blank(),
          panel.grid = element_blank()) +
    theme(panel.border = element_rect(fill=NA, colour = "darkgray", size=0.5)) +
    labs(x=NULL, y=NULL, title="Accumulated results [%]")

# plot_piechart

plot_dist_md <-
    ggdraw() +
    draw_plot(plot_barchart) +
    draw_plot(plot_piechart, 0.55, 0.55, 0.40, 0.40)

plot_dist_md

plot_dist_md_file_name <- "fig3_dist_md.png"
cowplot::ggsave2(plot = plot_dist_md, filename = plot_dist_md_file_name, device = "png", 
                 path = here::here("figs"), scale = 1, 
                 width = 22, height = 18, units = "cm", dpi = 600)

```


To complement the previous figure, some percentages are discussed in the body of the article which are included in the following table. Note: the table as it is does not appear on the paper.

```{r stats_assesstype_md, echo=FALSE}

data_plot_barchart %>%
    # arrange(desc(md_desc)) %>%
    select(`Mental Disorder` = md_desc,
           `Type of assessment` = val_asstype,
           `Percentage`= lbl) %>%
    knitr::kable(format="html", escape = T, booktabs = TRUE) %>%
    kable_styling(bootstrap_options = c("striped", "hover", "condensed", "responsive")) %>%
    column_spec(1, bold = T) %>%
    collapse_rows(columns = 1, valign = "top")
```


\newpage


### Figure 4: Distribution of articles published for the top 6 mental disorders over time.

A _line chart_ to show the number of papers per mental disorder and year. 


```{r fig4_linechart, echo=FALSE, dpi=600, fig.width=12, fig.height=14, fig.asp=0.65}
cols_top <- c("Depressive disorders", 
              "Anxiety disorders", 
              "Various disorders",
              "Trauma and\n stressor-related disorders",
              "Substance-related and\n addictive disorders",
              "Schizophrenia spectrum and\n other psychotic disorders")
top <- levels(all_data$md)[1:6]

data_plot_topmd <- 
    all_data %>%
      filter(md %in% top) %>%
      group_by(md_desc, year) %>%
      summarise(number_cases = n()) %>%
      ungroup() %>%
      mutate(md_desc = factor(md_desc)) %>%
      arrange(desc(number_cases))

#' Add mental disorders with zero cases
for (y in 2013:2018) {
    md_year <- filter(data_plot_topmd, year==y) %>% select(md_desc)
    md_dif <- setdiff(cols_top, md_year$md_desc)
    if (length(md_dif) > 0) {
        for (md in md_dif) {
            data_plot_topmd <- add_row(data_plot_topmd, md_desc=md, year=y, number_cases=0)
        }
    }
}


data_plot_topmd$md_desc <- forcats::fct_relevel(data_plot_topmd$md_desc, cols_top)
brks <- levels(data_plot_topmd$md_desc) 


plot_topmd <-
    data_plot_topmd %>%
        ggplot(aes(x=year, y=number_cases, group=md_desc)) +
        geom_vline(aes(xintercept = 2017), color="lightgray", linetype = "dashed", size = 0.5) +
        annotate("rect", xmin = 2017, xmax = 2018, ymin = -Inf, ymax = +Inf, fill = "lightgray", alpha = 0.2) +
        geom_line(aes(color=md_desc), size=2.5, alpha=.4) +
        geom_point(shape=21, size=7, color="darkgray", fill="white") +
        labs(#title="Distribution top mental disorders per year", 
             #subtitle = paste0("Top mental disorders (", top_lbl,")"),
             x="Year", 
             y="# of papers", 
             caption="Source: authors") + 
        scale_color_brewer(name="Mental disorders", palette="Set2", breaks=brks) +
        # geom_text(aes(label = number_cases), color= "white", size=3) +
        geom_text(aes(label = number_cases), color= "black", size=3) +
        scale_y_continuous(breaks=seq(0,8,by=1)) +
        theme_minimal() +
        # Legend: Top-Left Inside the Plot") 
        theme(legend.justification = c('left', 'top'),
              legend.position=c(0.05, 0.95),  
              legend.background = element_rect(color = "darkgray", size = 0.5, linetype ="solid"),
              legend.key = element_blank()) +
        theme(panel.grid.major = element_blank()) + 
        theme(panel.grid.minor = element_blank()) +
        theme(panel.background = element_blank()) +
        theme(plot.margin=unit(rep(20, 4), "pt")) +
        # Change the line type and color of axis lines
        theme(axis.line = element_line(colour = "darkgray", size = 0.5, linetype = "solid"))

plot_topmd
plot_file_name <- "fig4_linechart.png"
cowplot::ggsave2(plot = plot_topmd, filename = plot_file_name, device = "png", 
                 path = here::here("figs"), scale = 1, 
                 width = 20, height = 18, units = "cm", dpi = 600) 

```


\newpage


### Table 1: Mental disorders and the studies targeting them: NA = app name not available/not mentioned.

A tabular, compact distribution of papers (apps) per mental disorder, grouping the references per app. The number(s) in brackets next to the app name is the reference(s) in which the app is mentioned. In the table below, these references are internal identifiers instead. 


```{r tab1, echo=FALSE}

unite_paper_ids <- function(mentaldisorder, appname) {
    if (!is.na(appname)) {
        all_data %>% 
            filter(md_desc == mentaldisorder & app_name==appname) %>%
            arrange(year) %>%
            select(id) %>%
            as_vector() %>%
            stringr::str_c(collapse = ";")    
    } else {
         all_data %>% 
            filter(md_desc == mentaldisorder & is.na(app_name)) %>%
            arrange(year) %>%
            select(id) %>%
            as_vector() %>%
            stringr::str_c(collapse = ";")  
    }
}

data_kp_apps <- 
    all_data %>%
    group_by(md_desc, app_name) %>%
    summarise(number_apps = n()) %>% 
    arrange(number_apps, md_desc)

data_kp_apps <- 
    data_kp_apps %>%
    add_column(ids = purrr::map2(data_kp_apps$md_desc, data_kp_apps$app_name, unite_paper_ids))

data_kp_apps <- 
    data_kp_apps %>%
    add_column(app_ids = paste0(data_kp_apps$app_name, " (", data_kp_apps$ids, ")"))

data_kp_apps <- 
    data_kp_apps %>%
    group_by(md_desc) %>%
    summarise(app_ids_merge = paste0(app_ids, collapse = ", ")) 

data_kp_apps$md_desc <- factor(data_kp_apps$md_desc, levels=md_all_ordered)  


# options(knitr.kable.NA = '-')
data_kp_apps %>%
    arrange(desc(md_desc)) %>%
    select(`Mental Disorder` = md_desc,
           `References by app` = app_ids_merge) %>%
    knitr::kable(format="html", escape = T, booktabs = TRUE,    
          caption = paste0("Compact distribution of papers (apps) per mental disorder\n", 
                            "'NA' is app not specified/available")) %>%
    kable_styling(bootstrap_options = c("striped", "hover", "condensed", "responsive")) %>%
    column_spec(1, bold = T) %>%
    collapse_rows(columns = 1, valign = "top") 

```

\newpage


### Figure 5: Bubble plot representing technology-related dimensions (software features - orange; built-in sensors - green; analytics - blue) versus mental disorders. Bubble size corresponds with amount of articles.

It is a _bubble plot_ that shows the distribution of papers per mental disorder (bubble size) and technology-related characteristics grouped by dimensions (bubble color): software features, built-in sensors, and analytics. Technology-related characteristics are ranked in each dimension.


```{r fig5_feat, echo=FALSE, warning=FALSE}
cols = c("id", "md_id", "md_desc", 
         "feat_use", "feat_promp", "feat_soc", "feat_hcp", "feat_learn", 
         "feat_prog", "feat_ca", "feat_ass", "feat_vr", "feat_ar", "feat_pers", "feat_game",
         "app_name", "year")

cols_feat = c("feat_use", "feat_promp", "feat_soc", "feat_hcp", "feat_learn", 
              "feat_prog", "feat_ca", "feat_ass", "feat_vr", "feat_ar", "feat_pers", "feat_game")

data_kp_feat <- 
    all_data %>%
    select(cols) %>%
    gather(cols_feat, key="tech_type",value="tech_value") %>%
    mutate(cat = "Software features") %>%
    filter(tech_value == "YES") %>%
    group_by(md_desc, tech_type, cat)  %>%
    summarise(number_cases = n())  %>%
    ungroup() # required to add_row() 


# For totals in final bubble plot
data_kp_feat_n <-
    data_kp_feat %>%
    group_by(tech_type) %>%
    summarise(n = sum(number_cases)) %>%
    arrange(desc(n))
                  

md_feat <- unique(data_kp_feat$md_desc)
md_dif <- setdiff(md_all_ordered, md_feat)
if (length(md_feat) > 0) {
    for (md in md_dif) {
        # Added "feat_use" (or any value) to avoid NA in the "tech_type" variable. Nothing is drawn in the plot
        data_kp_feat <- add_row(data_kp_feat, md_desc=md, tech_type="feat_use", cat="Software features")
    }
}

data_kp_feat$md_desc <- factor(data_kp_feat$md_desc, levels=md_all_ordered)  

```

```{r fig5_sens, echo=FALSE, warning=FALSE}
cols = c("id", "md_id", "md_desc", 
         "sens_acc", "sens_gyr", "sens_gps", "sens_mic", "sens_cam",
         "app_name", "year")

cols_sens <- c("sens_acc", "sens_gyr", "sens_gps", "sens_mic", "sens_cam")
data_kp_sens <-
    all_data %>%
    select(cols) %>%
    gather(cols_sens, key="tech_type", value="tech_value") %>%
    mutate(cat = "Built-In sensors") %>%
    filter(tech_value == "YES") %>%
    group_by(md_desc, tech_type, cat) %>%
    summarise(number_cases = n()) %>%
    ungroup() # required to add_row() 


# For totals in final bubble plot
data_kp_sens_n <-
    data_kp_sens %>%
    group_by(tech_type) %>%
    summarise(n = sum(number_cases)) %>%
    arrange(desc(n))
                  

md_sens <- unique(data_kp_sens$md_desc)
md_dif <- setdiff(md_all_ordered, md_sens)
if (length(md_dif) > 0) {
    for (md in md_dif) {
        # Added "sens_mic" (or any other values) to avoid NA in the "tech_type" variable. Nothing is drawn in the plot
        data_kp_sens <- add_row(data_kp_sens, md_desc=md, tech_type="sens_mic", cat="Built-In sensors")
    }
}

data_kp_sens$md_desc <- forcats::fct_relevel(data_kp_sens$md_desc, levels=md_all_ordered)


```



```{r fig5_anal, echo=FALSE, warning=FALSE}
cols = c("id", "md", "md_id", "md_desc", 
         "anal_ml", "anal_beh", "anal_act", "anal_sp", 
         "app_name", "year")

cols_anal = c("anal_ml", "anal_beh", "anal_act", "anal_sp")

data_kp_anal <- 
    all_data %>%
    select(cols) %>%
    gather(cols_anal, key="tech_type",value="tech_value") %>%
    mutate(cat = "Analytics") %>%
    filter(tech_value == "YES") %>%
    group_by(md_desc, tech_type, cat)  %>%
    summarise(number_cases = n())  %>%
    ungroup() # required to add_row() 

# For totals in final bubble plot
data_kp_anal_n <-
    data_kp_anal %>%
    group_by(tech_type) %>%
    summarise(n = sum(number_cases)) %>%
    arrange(desc(n))

md_anal <- unique(data_kp_anal$md_desc)
md_dif <- setdiff(md_all_ordered, md_anal)
if (length(md_anal) > 0) {
    for (md in md_dif) {
        # Added "anal_ml" (or any value) to avoid NA in the "tech_type" variable. Nothing is drawn in the plot
        data_kp_anal <- add_row(data_kp_anal, md_desc=md, tech_type="anal_ml", cat="Analytics")
    }
}

data_kp_anal$md_desc <- factor(data_kp_anal$md_desc, levels=md_all_ordered)  

```


```{r fig5_alltogether, echo=FALSE, warning=FALSE, dpi=600, fig.width=15, fig.height=14, fig.asp=0.65}

data_kp_all <-
    bind_rows(data_kp_feat, data_kp_sens, data_kp_anal)

data_kp_all$tech_type <- as_factor(data_kp_all$tech_type)
#' Replace existing factors levels for figure production
levels(data_kp_all$tech_type)[levels(data_kp_all$tech_type)=="feat_learn"] <- "Learning"
levels(data_kp_all$tech_type)[levels(data_kp_all$tech_type)=="feat_prog"] <- "Progress"
levels(data_kp_all$tech_type)[levels(data_kp_all$tech_type)=="feat_pers"] <- "Personalization"
levels(data_kp_all$tech_type)[levels(data_kp_all$tech_type)=="feat_ass"] <- "Assessment"
levels(data_kp_all$tech_type)[levels(data_kp_all$tech_type)=="feat_promp"] <- "Prompting"
levels(data_kp_all$tech_type)[levels(data_kp_all$tech_type)=="feat_hcp"] <- "Health Care Provider \nCommunication"
levels(data_kp_all$tech_type)[levels(data_kp_all$tech_type)=="feat_use"] <- "In-Situ Use"
levels(data_kp_all$tech_type)[levels(data_kp_all$tech_type)=="feat_soc"] <- "Social"
levels(data_kp_all$tech_type)[levels(data_kp_all$tech_type)=="feat_ca"] <- "Context-Awareness"
levels(data_kp_all$tech_type)[levels(data_kp_all$tech_type)=="feat_game"] <- "Game"
levels(data_kp_all$tech_type)[levels(data_kp_all$tech_type)=="feat_vr"] <- "Virtual Reality"
levels(data_kp_all$tech_type)[levels(data_kp_all$tech_type)=="feat_ar"] <- "Augmented Reality"

levels(data_kp_all$tech_type)[levels(data_kp_all$tech_type)=="sens_mic"] <- "Microphone"
levels(data_kp_all$tech_type)[levels(data_kp_all$tech_type)=="sens_gps"] <- "GPS"
levels(data_kp_all$tech_type)[levels(data_kp_all$tech_type)=="sens_cam"] <- "Camera"
levels(data_kp_all$tech_type)[levels(data_kp_all$tech_type)=="sens_acc"] <- "Accelerometer"
levels(data_kp_all$tech_type)[levels(data_kp_all$tech_type)=="sens_gyr"] <- "Gyroscope"

levels(data_kp_all$tech_type)[levels(data_kp_all$tech_type)=="anal_ml"] <- "Machine Learning"
levels(data_kp_all$tech_type)[levels(data_kp_all$tech_type)=="anal_act"] <- "Activity Analysis"
levels(data_kp_all$tech_type)[levels(data_kp_all$tech_type)=="anal_beh"] <- "Behavioral Analysis"
levels(data_kp_all$tech_type)[levels(data_kp_all$tech_type)=="anal_sp"] <- "Spatial Analysis"

cols_ordered <-  c("Learning", "Progress", "Personalization", "Assessment", "Prompting", "Health Care Provider \nCommunication", 
                   "In-Situ Use", "Social", "Context-Awareness", "Game",  "Virtual Reality", "Augmented Reality", 
                   "Microphone", "GPS", "Camera", "Accelerometer", "Gyroscope",  
                   "Machine Learning", "Activity Analysis", "Behavioral Analysis", "Spatial Analysis") 
dims_ordered <- c("Software features", "Built-In sensors", "Analytics")

data_kp_all$tech_type <- forcats::fct_relevel(data_kp_all$tech_type, cols_ordered)

data_kp_all_n <-
    bind_rows(data_kp_feat_n, data_kp_sens_n, data_kp_anal_n)

data_kp_all_n <-
    data_kp_all_n %>%
    mutate(tech_label = cols_ordered)


data_kp_all$cat <- factor(data_kp_all$cat)
data_kp_all$cat <- forcats::fct_relevel(data_kp_all$cat, dims_ordered)
brks <- levels(factor(data_kp_all$cat))

# Color per dimension: 
# Software features "#FC8D62"; Built-In sensors "#66C2A5"; Types of analyses "#8DA0CB"
lbls_colors <- 
    ifelse(unique(data_kp_all$tech_type) %in% c("Learning", "Progress", "Personalization", "Assessment", "Prompting", "Health Care Provider \nCommunication", "In-Situ Use", "Social", "Context-Awareness", "Game",  "Virtual Reality", "Augmented Reality"), "#FC8D62",
           ifelse(unique(data_kp_all$tech_type) %in% c("Microphone", "GPS", "Camera", "Accelerometer", "Gyroscope"), "#66C2A5",
           "#8DA0CB"))
               
kp_bubblechart <- 
    data_kp_all %>%
    ggplot(aes(x=md_desc, y=tech_type, colour=cat)) +
    geom_point(aes(size=number_cases), alpha=1, na.rm = TRUE)+#, show.legend = FALSE) +
    geom_text(aes(label=number_cases), colour="black", size=3, na.rm = TRUE) +
    scale_size_area(max_size=18) +
    coord_flip() +
    labs(#title="Mental disorders vs Technology", 
         #subtitle = "Technology-related characteristics are ranked in each dimension",
         x="Mental disorders", 
         y="Technology-related characteristics", 
         caption="Source: authors") + 
    scale_color_manual(name="Dimensions", breaks=brks,
                       values = c("Software features"="#FC8D62","Built-In sensors"="#66C2A5", "Analytics"="#8DA0CB")) +
    # Which legend to show
    guides(colour="legend",size = "none") +
    theme_minimal()  +
    theme(axis.text.x=element_text(angle=60, size=11, hjust=1, color=lbls_colors),
          axis.text.y=element_text(size=11)) +
    # Legend: Top-Right Inside the Plot") 
    theme(legend.title = element_text(size=9),
          legend.justification = c('right', 'top'),
          legend.position=c(1, 0.90),
          legend.background = element_rect(color = "darkgray", size = 0.5, linetype ="solid"),
          legend.key = element_blank()) +
    # Change the line type and color of axis lines
    theme(axis.line = element_line(colour = "darkgray", size = 0.5, linetype = "solid")) +
    theme(panel.grid.major.x = element_blank()) + 
    theme(panel.grid.minor = element_blank()) +
    theme(panel.background = element_blank()) +
    theme(plot.margin=unit(rep(20, 4), "pt")) 

# Add annotations: total of columns
kp_bubblechart <-
    kp_bubblechart +
    annotate("rect", xmin = 14.40, xmax = 14.75, ymin = 0.5, ymax = 12.4, 
             fill = "#FC8D62", alpha = 0.6) +
    annotate("rect", xmin = 14.40, xmax = 14.75, ymin = 12.6, ymax = 17.4, 
             fill = "#66C2A5", alpha = 0.6) +
    annotate("rect", xmin = 14.40, xmax = 14.75, ymin = 17.6, ymax = 21.5, 
             fill = "#8DA0CB", alpha = 0.6)
    
for (i in 1:nrow(data_kp_all_n)) {
     kp_bubblechart <-
        kp_bubblechart +
        annotate("text", x = "Personality disorders", y = data_kp_all_n$tech_label[i], label = data_kp_all_n$n[i],
                 color = "white", fontface="bold", size=3, hjust = 0.4, vjust = -3)
}

# Add rectangles to highlight group of bubbles
kp_bubblechart <-
    kp_bubblechart +
    annotate("rect", xmin = 0.5, xmax = 11.5, ymin = 0.5, ymax = 7.5, 
             fill = "lightgray", alpha = 0.2, linetype="dashed") +
    annotate("rect", xmin = 0.5, xmax = 5.5, ymin = 0.5, ymax = 9.5, 
             fill = "lightgray", alpha = 0.2)

kp_bubblechart

kp_file_name <- "fig5_bubble.png"
cowplot::ggsave2(plot = kp_bubblechart, filename = kp_file_name, device = "png", 
                 path = here::here("figs"), scale = 1, 
                 width = 40, height = 25, units = "cm", dpi = 600)


```
\newpage

## License

This document is licensed under a [Creative Commons Attribution 4.0 International License](https://creativecommons.org/licenses/by/4.0/).

The code is licensed under the [Apache License 2.0](https://choosealicense.com/licenses/apache-2.0/).

The data used is licensed under a [Open Data Commons Attribution License](https://opendatacommons.org/licenses/by/).


## Runtime environment description

```{r session_info, echo=FALSE}
devtools::session_info(include_base = TRUE)
```