Chronic diseases

Diabetes Mellitus (algorithm)


Data files
LASAzdc1: first cohort of LASA-B, LASA-C, LASA-D, LASA-E, LASAF, LASAG
LASAzdc2: second cohort of LAS2B, LASAF, LASAG
(pdf)

Contact: Lisa van Zutphen

Background
The prevalence of diabetes mellitus (DM) is increasing with aging. According to data from The Dutch National Institute for Public Health and the Environment (RIVM), about 14% of men and women aged 65-74 years old were diagnosed with DM in The Netherlands 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).

DM algorithm

To increase reliability of the presence of DM in LASA, an algorithm (decision tree) was developed. In 2018, the diabetes mellitus algorithm was revised by Lisa van Zutphen (physician) and Renate de Jongh (physician internal medicine).

The former algorithm combined data on DM from the main interview (self-report of chronic diseases), the medical interview (inspections of medicine bottles) and the medical records of general practitioners. In the former algorithm,  a ‘yes’ on the general practitioners variable resulted in a definite diagnosis of DM. In case of ‘missing’ or ‘no’ on the general practitioners variable, information on other aspects (self-report of DM and use of antidiabetic medication) was combined in order to determine the presence of DM. This, however, resulted in some contra-intuitive classifications, such as ‘yes’ on use of antidiabetic medication resulting in a ‘possible’ DM diagnosis according to the algorithm. For this reason, the algorithm was revised.

Major changes:


Data sources
Data files and variables used in the algorithm:

For detailed information on the questionnaires used to retrieve the data mentioned above, please click on the names in bold.

Self-report of DM
Question *diabe01 stated ‘Do you have diabetes?’. At baseline (B) the response categories were: Missing / No / Yes. At the subsequent cycles, the response categories were extended to: No, never / No, (previous)diabe01 Yes / Yes, (previous)diabe01 No / Yes, (previous)diabe01 Yes. For each cycle, the two response categories starting with ‘No, ….’ were collapsed as ‘no’, and the two response categories starting with ‘Yes, ….’ were collapsed as ‘yes’. This results in yes / no / missing as response categories for every cycle.

Use of antidiabetic medication
The respondents were asked to bring in the medication with packaging that they used in the past two weeks. All the medication was noted. Based on ATC-codes, a variable was created to indicate if a respondent uses antidiabetic medication (oral antidiabetics or insulin, syntax available upon request). Response categories are yes / no / missing, see Medication use.

Medical records
In plm. 1994, 2000/2001 and 2005/2006, general practitioners (GP) were asked to fill in a questionnaire with specific questions on the medical history of respondents. The questions used in the algorithm are ‘Does the respondent have diabetes mellitus?’ (yes/no/missing) and ‘Year of diagnosis?’.

The year of diagnosis was combined with the year in which the interview with the respondent took place. By doing this, we were able to match the year of GP’s diagnosis with the year in which the LASA cycle took place. At every wave, the GP’s diagnosis was categorized as: yes / no / ever / missing, with ‘missing’ consisting of the respondents of whom the GP did not answer the first question (Does the respondent have diabetes mellitus?) and ‘ever’ consisting of the respondents of whom the GP did answer the first question positive (Does the respondent have diabetes mellitus?) but did not answer the second question (Year of diagnosis?). If diabetes mellitus was already present according to a previous GP questionnaire, the diabetes mellitus status was changed into yes at every subsequent wave (regardless of the year of diagnosis in the previous GP questionnaire(s) and regardless of the answer on the first question in the next GP questionnaire). For example, a respondent had diabetes according to the GP questionnaire in 2000 (disregarding the year of diagnosis) and this respondent has no diabetes according to the GP questionnaire in 2005/2006, the GP’s diagnosis will be YES at the 2001/2002 wave and subsequent waves.

Furthermore, in order to reduce the proportion of missingness, we made optimal use of the GP data. For respondents with a missing GP’s diagnosis on a specific wave, the data of subsequent GP questionnaires will be taken into account. For example, if a respondent did not have diabetes according to a subsequent GP questionnaire, the GP’s diagnosis will be ‘no’ on that specific previous wave. The rationale behind this is as follows. While the completion rates of the GP questionnaires are quite high (ranging between 66.9% and 85.7%), there is a subgroup of respondents of whom GP data is missing. There might be, however, data available from previous or subsequent GP questionnaires can give additional information on the presence of DM on other waves. As only the year of onset and not the year of recovery or curing was asked, it is not possible to determine if a respondent no longer has DM based on the GP questionnaire.

The DM algorithm
Combining information on these three (self-report, general practitioner’s diagnosis, medication) aspects of DM and non-response data, respondents are categorized in 6 groups: definite DM / possible DM / contradictory DM / no DM / missing / dropout. The use of antidiabetic medication indicates that a respondent definitely has diabetes. Not using antidiabetics however, does not mean the opposite (i.e. not having diabetes), since respondents could have applied lifestyle changes. For more information on the categorization, see the algorithm. In short, the ‘definite DM’ group consists of all respondents with either the use of antidiabetic medication and/or ‘yes’ on both self-report of DM and General practitioner’s diagnosis. The remaining group, consisting of all respondents with either ‘no’ or ‘missing’ on use of antidiabetic medication were categorized as follows. The ‘possible DM’ group consists of respondents with ‘yes’ on either self-report of DM or General practitioner’s diagnosis, with ‘missing’ on the remaining variable (i.e. not enough data available to make it a definite diagnosis). The ‘contradictory DM’ group is formed by respondents with a combination of ‘yes’ and ‘no’ on self-report of DM and General practitioner’s diagnosis. All respondents with ‘no’ on either self-report of DM or General practitioner’s diagnosis, with a missing value on the remaining variable are categorized as ‘no DM’. Respondents with no’ or ‘missing’ on use of antidiabetic medication, and ‘missing’ on both self-report of DM and General practitioner’s diagnosis were categorized as ‘missing’.

Longitudinal cleaning

Based on the questionnaires used in LASA (for respondents and for general practitioners) and the cleaning process, it is not possible to determine if respondents recover from DM. For this reason, a second variable was created, which was based on the output of the DM algorithm (thus on the same data). If a respondent had a definite diagnosis of DM according to the algorithm, the DM status of all subsequent waves was changed into ‘definite DM’. This was done regardless of the output of the algorithm on those waves, except for non-response. In those cases, the DM status of all subsequent waves was changed into ‘definite DM’ until the respondent was classified as drop-out.

The syntax of the DM algorithm can be found here.

Availability of information per wave1

After longitudinal cleaning

B

C

D

E

2B*

F

G

n=3107

n=3107

n=3107

n=3107

n=1002

n=4109

n=4109

Definite

212

171

187

168

70

240

238

Possible

26

19

12

17

5

14

18

Contradictory

65

52

35

36

7

47

38

No

2786

2302

1840

1464

917

1864

1521

Missing

18

1

2

6

3

0

3

Dropout

0

562

1031

1416

0

1944

2291

1 More information about the LASA data collection waves is available here.

* 2B=baseline second cohort

Previous use in LASA

Revised algorithm: no

Original algorithm:

  • Pijpers, E., Ferreira, I., De Jongh, R.T., Deeg, D.J.H., Lips, P.T.A., Stehouwer, C.D.A., Nieuwenhuijzen Kruseman, A.C. (2012). Older individuals with diabetes have an increased risk of recurrent falls: analysis of potential mediating factors: the Longitudinal Ageing Study Amsterdam Age and Ageing, 41, 358-365
  • Dik, M.G., Jonker, C., Comijs, H.C., Deeg, D.J.H., Kok, A., Yaffe, K., Penninx, B.W.J.H. (2007). Contribution of metabolic syndrome components to cognition in older individuals. Diabetes Care, 30, 10, 2655-2660.

References

  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.