A list of 23 common conditions was derived a priori from previous studies and from sources listing the most common chronic conditions in older adults.6 (link)–8 ,12 (link),22 (link) Conditions were selected that were common, managed with specific and reasonably defined drug therapy regimens, and identifiable using International Classification of Diseases, Ninth Revision (ICD-9) encounter codes. The Healthcare Cost and Utilization Project Clinical Classification Software was used to map ICD-9 diagnosis codes for each individual to each of the 23 conditions.23 These conditions included osteoarthritis and arthritis not otherwise specified (herein called arthritis); coronary heart disease (CHD); cerebrovascular accident (CVA); peripheral arterial disease (PAD; including aortic aneurysms); chronic obstructive pulmonary disease and asthma (COPD); diabetes mellitus; heart failure; atrial fibrillation; depression; hypertension; malignancy other than prostate cancer; prostate cancer; anemia; hypothyroidism; dementia; epilepsy; benign prostatic hypertrophy (BPH); Parkinson’s disease; osteoporosis; hyperlipidemia; gout; chronic renal insufficiency; and a combination category of gastroesophageal reflux disease, peptic ulcer disease, and dyspepsia (GERD/PUD). (Details of coding algorithms are provided in Appendix S1 .)
An individual was defined as having a condition of interest if he or she had one or more relevant ICD-9 codes during any outpatient visit or inpatient stay over the 2-year study period. Previous validation studies in the VA have identified this approach as highly sensitive and moderately to highly specific for identifying common chronic conditions.24 (link),25 (link) To maximize fidelity of outpatient diagnoses in VA data, codes were assessed only from encounter types in which a physician or nurse practitioner typically records diagnoses. (This approach does not consider diagnoses associated with radiology, laboratory, or other such visits.) In Medicare data, diagnoses were assessed only from claims arising from physician visits. In sensitivity analyses, a more-conservative method was used that required two or more ICD-9 codes from distinct outpatient visits or one or more codes from hospital discharge diagnoses to identify a disease condition as present.
An individual was defined as having a condition of interest if he or she had one or more relevant ICD-9 codes during any outpatient visit or inpatient stay over the 2-year study period. Previous validation studies in the VA have identified this approach as highly sensitive and moderately to highly specific for identifying common chronic conditions.24 (link),25 (link) To maximize fidelity of outpatient diagnoses in VA data, codes were assessed only from encounter types in which a physician or nurse practitioner typically records diagnoses. (This approach does not consider diagnoses associated with radiology, laboratory, or other such visits.) In Medicare data, diagnoses were assessed only from claims arising from physician visits. In sensitivity analyses, a more-conservative method was used that required two or more ICD-9 codes from distinct outpatient visits or one or more codes from hospital discharge diagnoses to identify a disease condition as present.
