We have studied the prevalence of comorbidity in the Stockholm EPR corpus containing almost 600,000 patients from 900 clinics using the ICD- 10 codes assigned to each patient record. The proportion of patients with a valid ICD-10 code was 83.0%, and 41.5% of these had at least one comorbidity. The most frequent comorbidity combination with type 2 diabetes was essential hypertension (43.1%). Our approach seems feasible for large scale analysis of diagnostic codes in EPR databases.
Today a large amount of electronic patient records (EPR) are produced, which
are rarely reused. The EPR systems are also more and more centralized
covering both several hospitals and clinics. In the Scandinavian countries we have a
unique social security number for all citizens following us from birth to death,
as well as from clinic to clinic. Diagnoses are often coded by ICD [
Comorbidity is the presence of no less than two distinct conditions in an
individual [
Starfield et al. [
As the population of elderly people is increasing in many countries the
prevelence of chronic diseases are expected to rise. Schellevis et al. [
Type 2 diabetes is defined as a chronic disease characterized by reduced
insulin sensitivity in target tissues [
Detecting comorbidity in a large population is of clinical interest due to the fact that it may reveal new information useful for cause of diseases as well as for new treatment strategies. The aim of this study is to analyze comorbidity in clinical hospital setting in Sweden using an EPR database, and to investigate comorbidity combinations with diabetes. 2
The tools used in this work1 are SQL query language and Java programming language. Fig. 1 shows a simple illustration for procedure of comorbidity analysis.
The Stockholm EPR (SEPR) corpus that we have access to contains almost
900 clinics with almost 600,000 patients, that are registered with their social
security number, gender, age, admission and discharge date of the patient as
well as the ICD code of the diagnosis encompassing the years 2006, 2007 and the
first half of 2008 [
The main issue in this work is to explore what comorbidity combinations each individual patient has in their medical history. Hence, the data of patient’s id, birth year, and diagnosis code are extracted from the SEPR database with a SQL query. In the extracted data there are 2,756,082 diagnosis records from 584,600 patients (=N).
The diagnosis codes in the extracted data are, however, not written in a unified way. For instance, the alphabet is written either in upper-case (e.g. A00) or in lower-case (e.g. a00), and different marks are used (e.g. A00-0, A00.0, A00,0, etc). Hence, the extracted codes need to be normalized. A code expression in upper-case without any marks (e.g. A000) is used in the analyses. Also, there are missing, or syntactically wrong and not to correct diagnosis codes (i.e. invalid codes) in the extracted diagnosis data. Examples of invalid codes are: AF063, KV9, KV˚A, NCU49, NFJ09, etc. There are 1,772,013 valid ICD-10 codes (13,450 different full level codes and 1,956 different level 3 codes) and 805,568 invalid ICD-10 codes in the extracted data.
Some patients have duplicate ICD-10 codes in their medical records (178,501 duplicate cases). This is presumably because these patients have received medical care in more than one institution. ICD-10 codes are wrong or missing in many cases. In our case the percentage of invalid ICD-10 codes in the extracted data is approximately 31%.
Only valid ICD-10 codes are used and converted to level 3 codes (i.e. truncating the first three characters), and other codes are not converted or interpreted by free text diagnoses. When converting the valid full level ICD-10 codes into level 3 codes, a patient can have the same ICD-10 code no less than twice (e.g. before converting A00, A001 and A009 ⇒ after converting A00, A00 and A00). Multiple ICD codes for a patient are therefore counted only once. Also, ICD-10 codes in chapters 19–21 are excluded from the frequency analysis because these codes include causes of diseases and other factors related to health care, and therefore not considered as diseases.
Creating comorbidity combinations for type 2 diabetes makes it possible to
focus the analysis on this diagnosis. Also, there are similar comorbidity analyses
in previous research which can be a methodological reference to this work as
mentioned in the background [
Out of the 584,600 (=N) patients the number of patients with a valid ICD-10 code was 485,271 (83.0%), and 242,435 (41.5%) had at least one comorbidity (see Table 1).
The total number of patients with type 2 diabetes were 14,162 (=n), out of which 13,487 (95.2%) had at least one comorbidity. The most frequent comorbidity combinations with type 2 diabetes were essential (primary) hypertension (43.1%), and heart failure (17.0%) (see Table 2). 4
In our study the proportion of patients with missing valid ICD-10 codes (17.0%)
was somewhat lower than expected. A study [
The comorbidity figures in our study (41.5%) were higher than for example in Westert et al. [15] showing about one fifth of patients with more than one chronic condition. However, these figures include all health care settings and are limited to chronic diagnoses.
The most frequent comorbidity combinations with type 2 diabetes in our study were essential hypertension (43.1%) and heart failure (17.0%). Our figures are in line with Caughey et al. [16] reporting about 51–53% comorbidity with hypertension, and 12–42% with cardiovascular disease. However, their study includes only eight chronic diagnoses. Similar comorbidity figures are also found in Finland in a population study by Reunanen et al. [17].
Our method to extract and present information needs to become more streamlined since today there are some manual steps that have to be carried out. The strength of our method is that we use continuously growing everyday clinical data enabling more detailed analysis. The main weakness is that there is the large proportion of patients without a valid ICD-10 code in EPR databases. One method to solve this problem is to match the free text in the patient record with the ICD-10 code’s textual description to populate the record with ICD-10 codes (see [14]).
We believe that in the future hospital management and clinical research will monitor and analyze the ICD-10 codes and also SNOMED-CT [18] to assess health care and also to predict future needs. 5
In Swedish hospital care the proportion of patients with a valid ICD-10 codes seemed to be fairly high (83.0%), and the comorbidity about 41.5%. The most frequent comorbidity combinations with type 2 diabetes were essential hypertension and heart failure. Our study raises questions on the quality and analysis of diagnosis coding in hospital settings. However, our approach seems feasible for large scale analysis of diagnostic codes in EPR databases. Our results on high rates for diabetes comorbidity may have implication for both health care planning and delivery.
In the future we plan to populate our data with more ICD-10 codes extracted from diagnosis expressions in the free text, this is similar to the approach described by Roque et al. [14].
Acknowledgments. We would like to thank Maria Skeppstedt for her invaluable help in posing smart SQL queries to extract data from the Stockholm EPR corpus. 13. Socialstyrelsen (The National Board of Health and Welfare, In Swedish): Diagnosgranskningar utof¨rda i Sverige 1997-2005 samtard˚ inof¨r granskning . Artikelnummer: 2006-131-30 (2006) 14. Roque, F.S., Jensen, P.B., Schmock, H., Andreatta, M., Hansen, T., Søeby, K., Braedkjaer, S., Juhl, A., Werge, T., Jensen, L.J., Brunak, S.: Using electronic patient records to discover disease correlations and stratify patient cohorts, submitted manuscript (2011) 15. Westert, G.P., Satariano, W.A., Schellevis, F.G., van den Bos, G.A.M.: Patterns of comorbidity and the use of health services in the Dutch population. Eur J Public Health. 11(4), 365–72 (2001) 16. Caughey, G.E., Vitry, A.I., Gilbert, A.L., Roughead, E.E.: Prevalence of comorbidity of chronic diseases in Australia. BMC Public Health. 221(8) (2008) 17. Reunanen, A., Kangas, T., Martikainen, J., Klaukka, T.: Nationwide survey of comorbidity, use, and costs of all medications in Finnish diabetic individuals. Diabetes Care. 23(9), 1265–71 (2000). 18. SNOMED Clinical Terms User Guide, http://www.ihtsdo.org