In growthcharts/bdsreader: Read Data from the Basisdataset Jeugdgezondheidszorg

knitr::opts_chunk$set(echo = TRUE)

Problem

On March 11, 2021, we received a report that showed implausible extreme D-scores for very young children. According to the client, this case does not stand on its own. The pattern also occurs for other children.

Example

The following code reproduces the problem.

library(bdsreader)
library(jsonlite)
suppressPackageStartupMessages(library(tidyr))
suppressPackageStartupMessages(library(dplyr))

fn  <- system.file("extdata", "bds_v1.0", "test", "test25.json", package = "jamesdemodata")
tgt <- suppressMessages(read_bds(fn, append_ddi = TRUE, schema = "bds_v1.0.json"))
tgt %>% 
  filter(yname == "dsc")

The value of 26 is far higher than we would expect. Also the DAZ value of 4.84 indicates an out-of-range value. Incidentically, also note that the D-score of 62.5 (SDS 3.56) at the age of 1.16 years is higher than expected.

Inspection

The 28 individual scores for this child and the corresponding item descriptions and difficulties are

itembank <- dscore::builtin_itembank %>% 
  filter(key == "dutch") %>% 
  select(item, tau, label)
data <- tgt %>% 
  filter(substr(yname, 1, 3) == "ddi") %>% 
  select(age, yname, y) %>% 
  rename(item = yname) %>% 
  dplyr::left_join(itembank, by = "item") %>% 
  print(n = Inf)  

There are several interesting things to observe:

1. Item ddigmd055 has a difficulty of 26, and is very difficult for babies aged four weeks. A positive score on this item dominates the D-score, resulting in an extreme score;
2. Items ddicmm032 (Maakt Gevarieerde Geluiden) and ddicmm035 (Reageert Op Mondeling Verzoek) have no difficulty estimate, so the D-score calculation ignores both responses;
3. Items ddigmd068, ddigmd168 and ddigmd268 all receive a positive score at 14 month. Only ddigmd068 is indicated for that age, and ddigmd168 and ddigmd268 should not be considered to be administered.

Open issues

All issues relate to problems in the mapping between BDS numbers and itembank

1. Is administration of ddigmd055 at the age of four week correct? Or does the evaluation perhaps depend on age, so it is actually easier at four weeks? In the latter case, the item needs to be split into separate items to obtain a valid D-score.
2. Should ddicmm032 and ddicmm035 be added?
3. Is this really one response on ddigmd068 only? If so, JAMES seems to fail separating BDS 986 according age.

Actions

1. Administration of ddigmd055 of happen at four different visits (4 weeks, 8 weeks, 13 week, 26 weeks). The evaluation depends on age, so effectively there are four different items, each with its own difficulty. We estimate the difficulty of the item from the SMOCC data, collected in 1989/1990. At that time, the Van Wiechenschema registered head lag at only the age of 26 weeks. Hence, we have no difficulty estimates for 4/8/13 weeks. As a fix, we split BDS (ddigmd055) into four items: ddigmd155 (4 weeks), ddigmd255 (8 weeks), ddigmd355 (13 weeks) and ddigmd055 (26 weeks) based on age. We define the age intervals between visits as (in days) [1, 42] (= 4 weeks), [43-102] (= 8 weeks), [103-146] (= 13 weeks) and [147-366] (= 26 weeks). In the future, we need tau's for the earlier occasions to obtain a more informative D-score estimate.
2. The procedure for item splitting of BDS 986 was incomplete. We now use age intervals [0.75, 1.75) (ddigmd068), [1.75, 2.5) (ddigmd168) and [2.5, 3.5) (ddigmd168).
3. The following items do not have a difficulty estimate, and hence are ignored in D-score estimation: ddigmd155, ddigmd255, ddigmd355, ddicmm032, ddicmm035.

growthcharts/bdsreader documentation built on Oct. 21, 2024, 4:39 a.m.