MORTON ET AL. RESPOND

We appreciate Acton and Bullock's concerns about diabetes case identification and tobacco use ascertainment bias in Indian Health Service (IHS) data, in which we demonstrated that patients with diabetes were smoking at significantly higher rates than were patients without diabetes.

Our diabetes case definition was determined by at least 1 entry of a physician-assigned diabetes International Classification of Diseases, Ninth Revision¹ code in the 5-year study period, based on 2 abnormal glucose values or symptoms. All diabetes cases were validated by a corresponding date of diagnosis, and no cases were missing this date. In a clinical setting, why would a diabetes diagnosis require a second-visit diagnosis for validation?

Acton and Bullock cite Solberg et al.,² who recommend at least 2 outpatient codes for diabetes in 1 year for case definition. However, Solberg used health plan and insurance claims data to construct this algorithm, not clinic visits and medical records data. Using clinic visit and medical records data for disease case definition, as in our report, is likely more accurate than using health plan claims data. Data from medical records are recorded at the time of the visit rather than being extracted and compiled from various health plan or employer databases, which may lead to many more sources of recording, translation, and compilation error.

Moreover, our overall diabetes prevalence, 10.3%, was similar to those found in other IHS reports, as stated. New analyses showed that 51.3% received a diabetes diagnosis before 1998, and the rates of diabetes diagnosis per year from 1998 to 2003 hovered around the 10.3% value. We believe that diabetes case misclassification error in our analyses was highly unlikely.

Acton questions the accuracy of our tobacco prevalence rates in diabetes patients, and points to several studies reporting smoking in diabetes patients using the IHS Diabetes Care and Outcomes Audit data. These audit data are a random sample of national IHS clinic medical records only for patients with diabetes, and no validation studies have been published on the method of data randomization and extraction when creating the final dataset.³ Our dataset included all visit data from 5 IHS sites and was not a random sample, making a strict comparison with audit data impractical.

Acton and Bullock state in their letter that, since 1986, IHS has a “longstanding emphasis on smoking ascertainment” and cite Mayfield et al.⁴ for support. However, in accordance with our conclusions that IHS needs to improve tobacco screening methods, Mayfield et al.⁴ found a very low rate of compliance (approximately 32%) in IHS clinics recording tobacco use at each visit. In addition, when tobacco use data are collected at an IHS clinic visit, there is no indication that ceremonial tobacco use is distinguished from regular recreational use, the same problem as in national datasets mentioned in our article. Therefore, tobacco use rates from IHS clinic data are possibly inflated, owing to inappropriate questions at clinic intake, and simultaneously underestimated because of a lack of recording compliance across clinics.

Acton and Bullock's letter cites several unpublished sources, personal communications, and new analyses run at Acton's request to cite smoking prevalence data in diabetes patients that are similar to our overall rate of 29.8%. These data sources help support our results, rather than refute our observations. One of Acton and Bullock's sources reported a 23.4% rate of smoking in diabetes patients in 2007 IHS audit data,⁴ another reported a 27% rate in the Strong Heart Study, and another reported 29.6% in Centers for Disease Control and Prevention's Behavioral Risk Factor Surveillance System data from 2005 to 2006. In Minnesota, Acton's last source, the smoking rate in diabetes patients was almost twice as high as that found in our data (53.1%). If our smoking rates in diabetes patients were inaccurate, they were perhaps underestimated.

What remains in question is smoking ascertainment bias among individuals without diabetes, about which Acton and Bullock appear unconcerned. As stated in our article, we believe it feasible that IHS diabetes patients are more likely to be asked if they smoke, thus possibly underestimating smoking prevalence in patients without diabetes. This likelihood is supported by Okuyemi et al.,⁵ who demonstrated significant provider tobacco use recording bias. We agree that, comparatively speaking, whether patients without diabetes are significantly less likely to smoke needs further investigation.

We want to reemphasize the higher rate of smoking in IHS diabetes patients in our data, as well as new data provided by Acton and Bullock, when compared with the general population rate of approximately 26%. Acton and Bullock also state that only 31.6% of diabetes patients identified as smokers in the IHS audit data were referred for cessation counseling. Why not 100%?

Lowering smoking prevalence in IHS diabetes patients leads to lower complication rates, and lowering smoking rates in patients without diabetes can prevent many new cases of chronic diseases. Improving tobacco assessment system-wide in IHS clinics for more culturally appropriate prevention and control would be of significant public health benefit for American Indian populations.

We suggest, as is possible with other national longitudinal survey or medical record datasets, such as the US Census, the Behavioral Risk Factor Surveillance System, the National Health and Nutrition Examination, and medicare and medicaid health data, that IHS open the “data dialogue,” and consider creating an access mechanism for qualified research scientists from a variety of disciplines to step up the publication of timely and useful analyses and results from IHS data.