Cancer risk in East Asian patients associated with acquired haemolytic anaemia: a nationwide population-based cohort study

The Taiwan National Health Insurance program has been a single-payer and universal insurance program since 1995 and has enrolled almost 99 % of the citizens of Taiwan in 2007. The Taiwan Ministry of Health and Welfare authorized the National Health Research Institutes (NHRI) to manage all registration files and claims data and to establish the National Health Insurance Research Database (NHIRD). The NHRI created a scrambled and anonymous identification number for each insured person for linking files and to protect the privacy of patients. The NHRI gives the permission to access the data for qualified researchers. The authors have published several population-based studies on the risk of cancer in various clinical settings using the NHIRD [11‐15]. This study was conducted after the approval by the Research Ethics Committee of the China Medical University, Taichung, Taiwan (CMU-REC-101-012). This study was performed in accordance with the ethical standards of the 1964 Declaration of Helsinki. The informed consent was waived by the Research Ethics Committee.

For the purpose of research, specific data subsets were constructed for more timely distribution. Three relevant data subsets were chosen for this population-based study, namely, the Registry for Beneficiaries which contained each insured individual’s registration data such as gender, date of birth, occupation and coverage period; the Inpatient Expenditures by Admission which included original claim data of all inpatients and finally, the Registry for Catastrophic Illness Patient Dataset (RCIPD), a unique subset of the NHIRD. Inclusion in the RCIPD required pathologic proof of malignancy, and when in doubt, the application would be examined by an independent haematologist/oncologist medical expense reviewer. Patients who satisfied the criteria for the RCIPD can benefit from a considerable reduction in out-of-pocket expenses for their cancer care throughout the country; this may also create a second check-point control from the patient side to prevent under-reporting of cancer occurrence in the RCIPD. In this research, the disease history was assembled from the Inpatient file. The disease diagnosis was recorded as per the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD–9-CM).

We organized a population-based retrospective cohort study to investigate the association between AHA and subsequent cancer risk. The flow chart of the study population selection is shown in Fig. 1. The AHA cohort consisted of patients with newly-diagnosed AHA (ICD-9-CM 283, from the inpatient records from 2000 to 2008. The index date was set at six months after (the first episode if more than one) the diagnosis of AHA was given at the hospital discharge. Patients with pre-existing malignancy before the index date were excluded. The AHA patients were separated into two sub-cohorts: a non-AIHA sub-cohort (ICD-9-CM codes, 283.1, 283.2 and 283.9) and an AIHA sub-cohort (ICD-9-CM code = 283.0). Because of the nature of an ICD-9-CM diagnosis categorization, a drug-induced haemolytic event in a patient with glucose-6-phosphate dehydrogenase (G6PD) deficiency would be coded as inherited haemolytic anaemia and thus was not included in the studied cohort of AHA.

The comparison cohort comprised individuals who had not been diagnosed with AHA or pre-existing cancer from 2000 to 2008. In order to increase the statistical power, for each AHA patient, we randomly selected 10 comparison persons from the general population frequency-matched by age (per 5 years), gender, urbanization of residency and index date. The index date of the comparison person was randomly matched by the same index year as that of the AHA case. The comparison cohort also had the same half year of lag observation time. We excluded all individuals with cancer diagnosed before the index date. The main parameter under consideration in this study was the incidence of developing cancer (ICD-9-CM 140–208, from the RCPID). The follow-up was terminated when cancer developed, or censored when the patient withdrew from the insurance, lost to follow-up or deceased, or on 31st December 2010 (Fig. 1).

The study also collected the co-morbidity history for each study subject as confounding factors. The co-morbidities before the index date included diabetes mellitus (DM, ICD-9-CM 250), alcohol use disorders (ALD, ICD-9-CM 265.2, 291, 303, 305.0, 357.5, 425.5, 535.3, 571.0, 571.1, 571.2, 571.3, 980.0 and V11.3), chronic kidney disease (CKD, ICD-9-CM 585), splenomegaly (ICD-9-CM 289.4), liver cirrhosis (ICD-9-CM 571.2, 571.5 and 571.6), hepatitis B virus infection (HBV, ICD-9-CM 070.2, 070.3 and V02.61) and hepatitis C virus infection (HCV, ICD-9-CM V02.62, 070.41, 070.44, 040.51 and 070.54) from the inpatient file and systemic lupus erythematosus (SLE, ICD-9-CM 710.0) and rheumatoid arthritis (RA, ICD-9-CM 714) from the RCPID. The urbanization level of residency was based on several index including population density (people/km²), and population ratio of different educational levels, population ratio of elderly, population ratio of people of agriculture workers and the number of physicians per 100,000 people [16]. We categorized the urbanization of residency into 4 levels. The level 1 indicated the highest urbanization level and the level 4+ meant the lowest level.

We compared distributions of age group, gender and co-morbidities and the mean and standard deviation (SD) for age between AHA and comparison cohorts. We calculated the overall incidence density rates of cancer for both cohorts, using the total number of cancer events divided by the total sum of follow-up years for each cohort. Cox proportional hazards regression analysis was used to estimate the AHA cohort to the comparison cohort hazard ratio (HR) and 95 % confidence interval. Multivariable Cox model was used to calculate the adjusted hazard ratio (aHR) and 95 % confidence interval including sex, age, urbanization of residency and all comorbidities in the model. Further data analysis calculated the incidence of individual cancer and the related AHA cohort to comparison cohort aHR for major cancers, including cancers of lung (ICD-9-CM 162), liver (ICD-9-CM 155), colorectal (ICD-9-CM 153 and 154), breast (ICD-9-CM 174, only in female), lymphatic and hematopoietic tissue(ICD-9-CM 200–208) and others. We also used Kaplan-Meier method to measure and plot the cumulative incidence for both cohorts and used log-rank test to examine the difference between the 2 cohorts. The proportional hazards assumption was not violated in the scaled Schoenfeld residuals test (p = 0.21). SAS 9.3 software (SAS Institute, Cary, NC, USA) was used to manage and analyse the data. The cumulative incidence curve was plotted by SPSS. The significant level was set at less than 0.05 for two-side testing of p-value.