Chronic diseases

Diabetes Mellitus (algorithm)


LASA-B, LASA-C, LASA-D, LASA-E

Contact: Laura_Schaap

Background
The prevalence of diabetes mellitus (DM) is increasing significantly with aging. According to data from RIVM, about 14% of men and women aged 65-74 years old were diagnosed with DM in 2003 (1). Because of the proportional increase of the aging population and the increasing number of older people with obesity, it is expected that the prevalence of DM will increase with 36% between 1993 and 2010 (2).
To increase reliability of the presence of DM, an algorithm (decision tree pdf) was developed using data on DM in the main interview (self report of chronic diseases), the medical interview (inspections of medicine bottles) and through the medical records of general practitioners.

Data sources

Medical records of general practitioners (GP)
1. GP data of 1992/1993 (LASABg01, D. Kriegsman, this file is currently under review by L. Schaap and will be released asap)
2. GP data of 2000/2001 (LASACg01, B. Schalk)
3. GP data of 2005/2006 (LASAEg01, L. Schaap)
In these data collections specific questions were asked about when diabetes was present and whether the diagnosis was confirmed by a specialist.

Self reported DM
LASA(B-C-D-E)035: chronic diseases
The variable xdiabe01 assesses the presence of DM. Only when the respondent answered this xdiabe01 as ‘yes’, follow-up questions were asked. Of the follow-up questions, the variable xdiabe04 assesses whether the respondent is being treated for DM by the general practitioner or a specialist.

Medication use
All prescribed drugs were registered by the medical interviewers. The drugs were pooled in specific groups per indication, e.g. antidiabetics (LASAx352).

Cross-sectional decisions
Question xdiabe01 stated ‘Do you have diabetes?’ At B the response categories were: Missing, No and Yes. At the C, D and E cycles the response categories were extended to:
o No, never
o No, (previous)diabe01 Yes
o Yes, (previous)diabe01 No
o Yes, (previousdiabe01 Yes
For each cycle, the two response categories starting with ‘No, ….’ were categorized as ‘No’, and the two response categories starting with ‘Yes, ….’ were categorized as ‘Yes’.

The cross-sectional DM categories created in the decision tree are Missing, No, Possible and Definite.

In the decision tree the medical records of GPs weigh heavier than the opinion of the respondents as it is unlikely for respondents to have diabetes without the GP being aware of it.
If the GP data indicate that a respondent has diabetes, the respondent is classified as ‘Definite DM’. If however, GP data are missing, self report becomes the most important source. Due to these decisions there are differences in classifications (No DM, possible DM, definite DM) between the branches starting with GP: No and GP: missing. This hierarchy was confirmed with a DM expert (G. Nijpels, GP).

Right branch
When the GP data indicate that a respondent does not have DM and this is confirmed by self report, the respondent is classified as ‘No DM’. When GP and self report are inconclusive (GP: no, self report: yes), but medication data confirm the presence of DM, the respondents are classified as ‘Definite DM’. However, when medication data on DM are missing or absent, the self-reported treatment by GP or specialist should give the final classification. When the self reported treatment is positively scored, the respondents are classified as ‘Possible DM’. When the self reported treatment is negative or missing, the respondents are classified as ‘No DM’ When the GP data indicate absence of DM and self-report is missing, the presence of DM medication resulted in a classification ‘Possible DM’, while missing data or absence of DM medication resulted in ‘No DM’.

Left branch
If GP data are missing, self report becomes the most important source. When a positive self report is confirmed by medication data, the respondent is classified as ‘Definite DM’. When self report is not confirmed by medication data (medication: No or Missing) and self reported treatment is negative, the respondents are coded as ‘No DM’. When self reported treatment is positive or missing, respondents are classified as ‘Possible DM’. When self report is No or Missing, and this is confirmed by medication data, the respondents are coded as ‘No DM’. When self report is No or Missing, but medication data indicate the presence of DM, respondents are scored as ‘Possible DM’.
When all sources are missing, respondents are coded as ‘Missing’.

Longitudinal decisions
Longitudinal decisions were based upon the assumption that once a respondent has DM, he or she is affected by this chronic disease for the rest of his/her life. However, when DM was present at the B or C cycle, but was absent at two subsequent cycles, respondents were scored as “No DM”.
Furthermore, if, for instance, DM was definitely present at the B cycle and according to the algorithm at cycle C not or possibly present, the value of the C cycle was recoded into ‘3’; meaning “definite DM at the B cycle and not/possibly present at the C cycle”. The same was done at the D and E cycles:
‘3’ at D cycle: “definite DM at B or C cycle, and not/possibly present at the D cycle”
‘3’ at E cycle: “definite DM at B, C or D cycle, and not/possibly present at the E cycle”.
For longitudinal analyses, it is recommended to further recode this value 3 into 1 (definite DM), but the individual researcher should make this decision.

Number of persons with/without DM at the cycles B, C, D and E before and after longitudinal data cleaning

Before longitudinal cleaning

After longitudinal cleaning

B

C

D

E

B

C

D

E

Missing

14

807

1235

1642

idem

807

1235

1642

No DM

2843

2101

1669

1306

-

2092

1668

1300

Definite DM

229

163

177

133

-

163

177

133

Possible DM

21

36

26

26

-

23

18

18

DM at B

-

-

-

-

-

21

-

-

DM at B or C

-

-

-

-

-

-

8

-

DM at B, C, or D

-

-

-

-

-

-

-

13

Reference List
(1) www.nationaalkompas.nl.
(2) Baan CA, Feskens EJ. Disease burden of diabetes mellitus type II in the Netherlands: incidence, prevalence and mortality. Ned Tijdschr Geneeskd 2001; 145(35):1681-1685.

*** Algorithm DM at B cycle ***

** b_gp_dia: medical records GP **
** b_diabe: self reported DM **
** bmed_dia: medication data **

if (b_gp_dia=1) bal_dia=1.
if ((missing(b_gp_dia)) and b_diabe=0 and bmed_dia=0) bal_dia=0.
if ((missing(b_gp_dia)) and b_diabe=0 and bmed_dia=1) bal_dia=2.
if ((missing(b_gp_dia)) and b_diabe=0 and (missing(bmed_dia))) bal_dia=0.
if ((missing(b_gp_dia)) and b_diabe=1 and bmed_dia=1) bal_dia=1.
if ((missing(b_gp_dia)) and b_diabe=1 and bmed_dia=0 and (bdiabe04>=1 or (bdiabe04<0)) bal_dia=2.
if ((missing(b_gp_dia)) and b_diabe=1 and (missing(bmed_dia) and (bdiabe04>=1 or (bdiabe04<0))) bal_dia=2.
if ((missing(b_gp_dia)) and b_diabe=1 and (missing(bmed_dia) and (bdiabe04=0))) bal_dia=0.
if ((missing(b_gp_dia)) and b_diabe=1 and bmed_dia=0 and (bdiabe04=0)) bal_dia=0.
if ((missing(b_gp_dia)) and (missing(b_diabe)) and bmed_dia=0) bal_dia=0.
if ((missing(b_gp_dia)) and (missing(b_diabe)) and (missing(bmed_dia))) bal_dia=-1.
if ((missing(b_gp_dia)) and (missing(b_diabe)) and bmed_dia=1) bal_dia=2.
if (b_gp_dia=0 and b_diabe=0) bal_dia=0.
if (b_gp_dia=0 and b_diabe=1 and bmed_dia=1) bal_dia=1.
if (b_gp_dia=0 and b_diabe=1 and bmed_dia=0 and bdiabe04>=1) bal_dia=2.
if (b_gp_dia=0 and b_diabe=1 and bmed_dia=0 and (bdiabe04=0)) bal_dia=0.
if (b_gp_dia=0 and b_diabe=1 and bmed_dia=0 and (missing(bdiabe04))) bal_dia=0.
if (b_gp_dia=0 and b_diabe=1 and (missing(bmed_dia) and bdiabe04>=1)) bal_dia=2.
if (b_gp_dia=0 and b_diabe=1 and (missing(bmed_dia) and bdiabe04=0)) bal_dia=0.
if (b_gp_dia=0 and b_diabe=1 and (missing(bmed_dia) and missing(bdiabe04))) bal_dia=0.
if (b_gp_dia=0 and (missing(b_diabe)) and bmed_dia=1) bal_dia=2.
if (b_gp_dia=0 and (missing(b_diabe)) and bmed_dia=0) bal_dia=0.
if (b_gp_dia=0 and (missing(b_diabe)) and (missing(bmed_dia))) bal_dia=0.

value label bal_dia -1 'missings' 0 'no diabetes' 1 'definite diabetes' 2 'possible diabetes' .
variable label bal_dia 'B algorithm: diabetes mellitus'.