Ethics Statement

This study was initiated after approved by the Institutional Review Board of Buddhist Dalin Tzu Chi General Hospital, Taiwan. Since all identifying personal information was stripped from the secondary files before analysis, the review board waived the requirement for written informed consent from the patients involved.

Database

The National Health Insurance (NHI) program, which provides compulsory universal health insurance, was implemented in Taiwan in 1995. It enrolls up to 99% of the Taiwanese population and contracts with 97% of all medical providers. The database contains comprehensive information on insured subjects, including gender, date of birth, residential or work area, dates of clinical visits, the International Classification of Diseases (Ninth Revision) Clinical Modification (ICD-9-CM) codes of diagnoses, details of prescriptions, expenditure amounts and outcome at hospital discharge (recovered, died, or transferred out). A random sample with 1,000,000 people based on the 2005 reimbursement data was established for public access, and the group did not differ statistically significantly from the larger cohort in age, gender or health care costs, as reported by the Taiwan National Health Research Institute. [10], [11] The sampled group was used as our study cohort.

To avoid financial issues for patients with major illnesses, the NHI specifies 31 categories of catastrophic illness (e.g., dementia, cancers, chronic renal failure, etc.) that are exempt from co-payment. The attending physician of a patient diagnosed as catastrophic illness can submit related information in application for a catastrophic illness certificate (CIC). A committee formally reviews applications, and if approved, patients are then exempted from co-payment. [12].

Study Population

The sampled population has been followed from January 1, 2002 to December 31, 2009 (a total of eight years). First, We identified people older than 18 years who were still alive in 2005 as our study cohort. Mild traumatic brain injury (mTBI) was defined by ICD-9-CM code head concussion (850.0, 850.1, 850.5, or 850.9), intracranial injury of other and unspecified nature (854.0), or head injury, unspecified (959.01). [13] Dementia was defined by ICD-9-CM code 290 or 331.0 (Alzheimer’s disease) in any position of the diagnoses. To maximize case ascertainment, only 1) patients registered with dementia in the CIC or 2) those prescribed with either cholinesterase inhibitors (rivastigmine, donepezil and galantamine) or NMDA receptor antagonist (memantine) for dementia were included. Second, We excluded all patients with mTBI and dementia diagnosed before January 1, 2005. To ensure that exposed patients were not misclassified, we also excluded patients who had ever hospitalized with traumatic brain injuries to ensure the enrolled patients with mTBI were discharged directly after visits. After exclusion, we identified 28551 people with mTBI and 692382 without mTBI. Each person was tracked for 5 years from January 1, 2005 till December 31, 2009 to identify if he or she was diagnosed with dementia. These patients were then linked to the administrative data for the period 2005–2009 to calculate disease-free survival time, with cases censored for patients who drew back guarantees from the NHI Program or were still robust at end of follow-up (Figure 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. Flow diagram of the population-based study.

https://doi.org/10.1371/journal.pone.0062422.g001

Covariates

To better understand the risk of mTBI on dementia, several covariates were included for analysis. First, patient demographics included age, gender, urbanization level (urban, suburban, and rural areas) and socio-economic status (SES). The insurance enrollee category was used as a proxy measure of SES. Second, prevalence of selected comorbid conditions (diabetes, hypertension, coronary artery disease, hyperlipidemia, history of alcohol intoxication, history of ischemic stroke and intracranial hemorrhage) and Charlson comorbidity index (CCI) were included. CCI is a scoring system that includes weighting factors on important concomitant diseases that has been validated for use with ICD-9-CM coded administrative databases. [14], [15].

Statistical Analysis

The SAS statistical package, version 9.2 (SAS Institute, Inc., Cary, NC) was used for data analysis. Pearson’s chi-square test was used for categorical variables. The Nelson-Aalen cumulative hazard estimates were plotted to show different trends between patients with mTBI and those without. Cox proportional hazard regression model was then used to calculate the hazard ratio of dementia for people with mTBI after adjustments for age, gender, urbanization level, SES, diabetes, hypertension, coronary artery disease, hyperlipidemia, history of ischemic stroke, history of intracranial hemorrhage and CCI.

We did a subgroup analysis based on people older than 65 years to check if the effects of mTBI on developing dementia are similar. We also evaluate the risk of dementia in patients who had been hospitalized because of TBI to see if there is ‘dose-response’ effect in the relation of TBI and dementia.

To further assess the robustness of our results, we did sensitivity analyses [16] to evaluate how large the effect of an unmeasured confounder would be needed to account for the results. A two-tailed P value of <0.05 was considered significant.

Limitations

First, our findings are generated from administrative data. The primary outcome of dementia was derived from ICD-9-CM code, which is good for insurance reimbursement rather than being the substitute of precise operative definition. The validity of the diagnosis, i.e. sensitivity, specificity and accuracy, cannot be assessed. To overcome this misclassification bias, we restricted our inclusion to patients who were registered in the CIC or those prescribed with medications solely for dementia, as there should be impossible for false positive cases on this side of disease spectrum. Epidemiologic study showed the prevalence of dementia in the elderly (older than 65 years old) in Taiwan ranged from 1.7% to 4.4%, [17] which is similar with our sampled data (1.55%; 2000 cases in a total of 129122 people). The prevalence in 5-year age bands from 65 to 84 years, and for those aged 85 years and older was 0.41%, 0.82%, 1.62%, 2.55% and 4.0%, respectively. Although frequencies were lower compared with the global report published by Cleusa et al, [18] the upward trends were the same. We acknowledged that dementia patients with mild symptoms would not be enrolled and this approach decreases the generalizability of study findings, but we had decided this a must trade-off.

Second, Coding errors are also common in mTBI. [13], [19] In Bazarian et al., [13] the sensitivity of ICD-9-CM codes for mTBI was 45.9% with a specificity of 97.8%. In other words, people in the mTBI group are highly possible to have mTBI, while some people in the unexposed group may still have mTBI during the study period but the inclusion strategy failed to identify them. In this situation, the predicted effect of mTBI on dementia should be toward the null, but we still found significant risk of developing dementia in patients with mTBI.

Third, we could not obtain the clinical information of patients with mTBI, such as the Glasgow Coma Scale, findings on computed tomography of head, the injury mechanism and the initial presentations. By definition, labeling our cases as ‘mTBI’ may be inappropriate. However, it has been validated that the ICD9-CM codes have high specificity regarding diagnosis of mTBI. [13] Furthermore, we excluded those who were admitted to the hospitals to make sure the patients enrolled are really “mild”. Based on our inclusion criteria, although not totally precise, we think the cases in our study group are highly correlated with the definition of mTBI.

Fourth, the average duration from TBI to the diagnosis of dementia is short and we admitted that reverse causation (i.e., undiagnosed dementia resulted in TBI) might also exist. However, we excluded all dementia patients in the first 3 years of our study period to eliminate the effect as possible. Of note, patients in TBI group are older; it is possible the neurodegenerative impact after TBI could have stronger effect on the elderly. Furthermore, we found higher HR of dementia (4.44) in patients admitted with TBI and the reverse causation could not fully explain the ‘dose-response’ of injury severity.

In the last, although we extensively adjust for possible comorbidities, unmeasured cofounding is still an issue. Based on the nature of our dataset, we cannot take some important risk factors of dementia such as gene or family histories into account. However, these risk factors are unlikely associated with mTBI and it is reasonable that these are not confounders in our study. [20] Furthermore, the adjusted HR is significant enough that the residual confounding may not be able to fully explain the result.