Back to Journals » Clinical Epidemiology » Volume 10

Comparing causes of death of Hodgkin lymphoma and breast cancer patients between medical records and cause-of-death statistics

Authors de Vries S, Schaapveld M , Kardaun JWPF, de Bruin KH, Krol ADG , Lugtenburg PJ, Jacobse JN, Aleman BMP , van Leeuwen FE

Received 3 January 2018

Accepted for publication 25 June 2018

Published 16 October 2018 Volume 2018:10 Pages 1523—1531

DOI https://doi.org/10.2147/CLEP.S161359

Checked for plagiarism Yes

Review by Single anonymous peer review

Peer reviewer comments 3

Editor who approved publication: Professor Henrik Sørensen



Simone de Vries,1 Michael Schaapveld,1 Jan WPF Kardaun,2,3 Kim H de Bruin,2 Augustinus DG Krol,4 Pieternella J Lugtenburg,5 Judy N Jacobse,1 Berthe MP Aleman,6 Flora E van Leeuwen1

1Department of Epidemiology, The Netherlands Cancer Institute, Amsterdam, the Netherlands; 2Department of Health and Care, Statistics Netherlands, The Hague, the Netherlands; 3Department of Public Health, Academic Medical Center, Amsterdam, the Netherlands; 4Department of Radiotherapy, Leiden University Medical Center, Leiden, the Netherlands; 5Department of Hematology, Erasmus MC Cancer Institute, Rotterdam, the Netherlands; 6Department of Radiation Oncology, The Netherlands Cancer Institute, Amsterdam, the Netherlands

Objective: Obtaining accurate data about causes of death may be difficult in patients with a complicated disease history, including cancer survivors. This study compared causes of death derived from medical records (CODMR) with causes of death derived from death certificates (CODDC) as processed by Statistics Netherlands of patients primarily treated for Hodgkin lymphoma (HL) or breast cancer (BC).
Methods: Two hospital-based cohorts comprising 1,215 HL patients who died in the period 1980–2013 and 714 BC patients who died in the period 2000–2013 were linked with cause-of-death statistics files. The level of agreement was assessed for common underlying causes of death using Cohen’s kappa, and original death certificates were reviewed when CODDC and CODMR showed discrepancies. We examined the influence of using CODDC or CODMR on standardized mortality ratio (SMR) estimates.
Results: Agreement for the most common causes of death, including selected malignant neoplasms and circulatory and respiratory diseases, was 81% for HL patients and 97% for BC patients. HL was more often reported as CODDC (CODDC=33.1% vs. CODMR=23.2%), whereas circulatory disease (CODDC=15.6% vs. CODMR=20.9%) or other diseases potentially related to HL treatment were more often reported as CODMR. Compared to SMRs based on CODDC, SMRs based on CODMR complemented with CODDC were lower for HL and higher for circulatory disease.
Conclusion: Overall, we observed high levels of agreement between CODMR and CODDC for common causes of death in HL and BC patients. Observed discrepancies between CODMR and CODDC frequently occurred in the presence of late effects of treatment for HL.

Keywords: cause of death, Hodgkin lymphoma, breast cancer, mortality statistics

Introduction

In epidemiology, differences in mortality risks between patient cohorts and the general population are usually quantified using standardized mortality ratios (SMRs). SMR estimation requires the availability of age-, sex-, and period-specific mortality rates for the general population. Often, these mortality rates are available from nationwide registries. It is therefore important that mortality data from official death certificates reflect similar underlying cause-of-death patterns as cause-of-death information from medical records. Inconsistencies in processing of causes of death may have implications for the interpretation of SMRs in epidemiologic studies. Only few studies compared cause-of-death information from medical records or hospital discharge records with death certificates.15

In recent years, it has become challenging to obtain cause-of-death information in cancer survivor cohorts, especially in case of mortality long after initial cancer treatment. Patients are more often cured and discharged, may have changed addresses, and may have died in another hospital, hospice or at home. Furthermore, due to the increasing administrative burden on general practitioners (GPs), response rates of clinical follow-up questionnaires have dropped considerably. To complete information on causes of death, linkage with a national cause-of-death registry can be considered. However, accurate registration of causes of death may be more difficult in patients with a long interval between disease diagnosis and date of death or a complicated disease history. For example, Hodgkin lymphoma (HL) patients are generally young at the time of treatment and have an excellent prognosis. Yet, they have an increased risk to die from various late adverse events related to treatment, including among others, second malignancies, cardiovascular diseases, and infections.6,7 These late complications may affect cause-of-death coding in these patients. In contrast, breast cancer (BC) patients also have a good prognosis but a lower risk of adverse effects of treatment,8 which may lead to less inconsistency in cause-of-death coding. For the interpretation of epidemiological studies on cause-specific mortality, it is important to gain insight into the effects of using cause-of-death information from different sources in different populations of cancer survivors.

In this study, we therefore compared causes of death recorded in the medical files with cause-of-death information from death certificates as processed by the Dutch cause-of-death statistics (Statistics Netherlands) in patients treated for HL or BC. We also examined to what extent differences in classification of causes of death influence SMRs.

Methods

We used data from a hospital-based cohort of HL patients (n=3,019), treated in the Antoni van Leeuwenhoek hospital, Erasmus MC Cancer Institute or Leiden University Medical Center, between 1965 and 2000, before the age of 51 years, and data from a hospital-based cohort of female BC patients (n=5,600) treated in the Antoni van Leeuwenhoek hospital for stages I–III invasive BC between 2000 and 2009, before the age of 61 years. Patient selection and data collection have been described in detail previously.6,911,14 For the comparison of causes of death, we considered all 1,215 HL patients who died in the period 1980–2013 and all 714 BC patients who died in the period 2000–2013.

Causes of death for both HL and BC patients were obtained from hospital medical records by trained abstractors. If abstractors had difficulties in assigning a cause of death, the patient was discussed within the project team (MS, BMPA, FEvL, SdV, JNJ) until agreement was reached. If the cause of death could not be retrieved from the hospital charts, a questionnaire was sent to the patient’s GP regarding the patient’s medical history and direct and underlying causes of death. Response rates from the GPs were 94% in 200411,12 and 83% in 2012 for the HL cohort13 and 71% in 2016 for the BC cohort.14 Exact dates of death were confirmed through linkage with the Central Office of Genealogy in order to allow deterministic linkage of all cohort members with Statistics Netherlands. This study was exempt from the approval of institutional review board according to Dutch law because existing data from medical files were used, which was confirmed by the institutional review board of the Netherlands Cancer Institute (PTC17.0788). Questionnaires obtained from the GPs were anonymized for the researchers, as well as the data retrieved from the hospital medical records. All analyses were performed at Statistics Netherlands: no cause-of-death registry data were exchanged. Original death certificates at Statistics Netherlands were reviewed by SdV only under a confidentiality agreement.

In the Netherlands, the underlying cause of death for each decedent is registered by Statistics Netherlands. Attending physicians or medical examiners are obliged to fill out a death certificate, which is sent to Statistics Netherlands through the municipality where the death occurred. At Statistics Netherlands, the death certificates are coded and an underlying cause is selected according to the tenth revision of the ICD (ICD-10), with a maximum of four coded diseases per decedent. The ICD-10 defines the underlying cause of death as 1) the disease or injury that initiated the train of morbid events leading directly to death or 2) the circumstances of the accident or violence that produced the fatal injury, with Volume 2 of ICD-10 providing several instructions for the application of this definition.15 Detailed information on methods of registration of causes of death in the Netherlands has been published previously.16,17

The hospital-based cohorts of HL and BC patients were linked with the causes of death at Statistics Netherlands to enable comparison of the causes of death derived from the medical records (CODMR) with causes of death derived from the death certificates (CODDC). Individual-level linkage was based on sex, date of birth, date of death, and filing municipality (first linkage step). If linkage failed due to missing information on filing municipality in the cohort data, linkage was based on the remaining three variables (second linkage step). Additionally, we performed a third linkage procedure where we varied parts of the linkage key (eg, day of death or day of birth). By chance, it is possible that two individuals have an identical linkage key (0.05% of total on the first linkage step and 1.3% on the second linkage step in the period 1980–2013). However, we checked for duplicate matches in both cohorts on the first two linkage keys and found none. Linkage was successful for 1,140 deceased patients (93.8%) in the HL cohort and 690 deceased patients (96.6%) in the BC cohort. The difference in successful linkages can be understood when realizing that deaths within the HL cohort partly occurred in earlier decades.

Statistical analyses

We compared the CODMR and the CODDC for common causes of death according to the ICD-10, including malignant neoplasms and circulatory disease, for all patients who were successfully linked with Statistics Netherlands. The causes of death of patients for whom the cause of death in our cohort was missing (unable to retrieve from the medical records and no response from the GP), or unspecified (ICD-10 codes R96–R99) were categorized as unknown or unavailable, including for 373 deceased HL patients (32.2%) and 113 deceased BC patients (16.4%). We calculated the conditional agreement between the CODMR and the CODDC in patients for whom the cause of death was available from both sources, thus excluding patients with unknown or unavailable cause of death. Conditional agreement for medical records was defined as the number of deceased patients for whom CODMR was in agreement with CODDC divided by the total number of deaths from a specific cause according to CODMR and vice versa for the conditional agreement for death certificates. Overall agreement between CODMR and CODDC was calculated by dividing the number of patients for whom the causes of death from CODMR and CODDC agreed by the total number of deceased patients. To adjust for chance agreement, the Cohen’s kappa statistic was calculated. Kappa values of >0.75 were considered to represent excellent agreement, values of 0.40–0.75 represented moderate agreement, and values of <0.40 represented poor agreement.18

The original death certificates were reviewed when CODMR differed from CODDC. Discrepancies were examined when HL was coded as CODMR but not as CODDC, or vice versa, in the HL cohort. Similarly, discrepancies were further examined when BC was coded as CODMR and but not as CODDC, or vice versa, in the BC cohort. Additionally, discrepancies between CODMR and CODDC were investigated when in the HL cohort, both CODMR and CODDC concerned a common malignancy as cause of death (including lung cancer, gastrointestinal cancer, non-HL and leukemia) or when CODMR was recorded as a common malignancy and CODDC specified circulatory disease, or vice versa. For examination of such discrepancies, only original death certificates filed between 1980–1989 and 2000–2013 were accessible.

Finally, we assessed whether and to what extent discrepancies in causes of death influenced the results of person-year analyses of cause-specific mortality. We compared the observed mortality in our HL cohort and mortality in the general population, taking into account the person-years of observation of patients in our cohort by age, sex, and calendar period. Mortality data for the general population from Statistics Netherlands for the period 1980–2013 were used as reference rates. In this analysis, time at risk began at start of treatment and ended at date of death, date of emigration, or December 31, 2013, whichever came first. SMRs were calculated as the ratio of the observed and expected number of deaths in our HL cohort, and the confidence limits were calculated using exact Poisson probabilities of observed numbers. The absolute excess risk (AER) was calculated as the observed minus the expected number of deaths divided by person-years at risk, expressed per 10,000 person-years. First, we performed the analysis using CODMR supplemented by CODDC if CODMR was missing. Subsequently, we performed the analysis using only CODDC. Lastly, we performed the analysis in the same way as the first analysis, and we additionally supplemented CODMR with CODDC if the data between the two sources showed discrepancy.

Results

In total, we included 1,140 deceased HL patients and 690 deceased BC patients for whom linkage with Statistics Netherlands could be performed. Table 1 shows the descriptive characteristics of both patient groups. The majority of deceased HL patients were male (61.8%). Median age at primary cancer diagnosis was 30.4 years (interquartile range [IQR]=23.7–38.6) for HL patients and 50.3 years (IQR=42.7–55.6) for BC patients. For HL patients, the median interval between HL treatment and death was 17.4 (IQR=8.0–27.0), and for BC patients, this interval was 4.3 years (IQR=2.4–6.8). Cause of death was available from both the medical records and death certificates for 767 HL patients and 577 BC patients.

Table 1 Descriptive characteristics of deceased HL and BC patients

Abbreviations: HL, Hodgkin lymphoma; BC, breast cancer; IQR, interquartile range.

The level of agreement between medical records and death certificates for common causes of death in HL patients (including HL; breast, lung, and gastrointestinal cancers; non-HL and leukemia; other malignant neoplasms; circulatory disease; and respiratory disease) is shown in Table 2. Using either CODMR or CODDC, the majority of patients died from a malignant neoplasm (66.5% and 75.9%, respectively). According to CODMR, 23.2% of all patients died from HL compared to 33.1% according to CODDC. When the CODDC stated HL, only in 63% of deaths, CODDC and CODMR agreed. The conditional agreement for HL as CODMR was 89.9%, while the corresponding Cohen’s kappa value of 0.64 indicated moderate agreement. On the other hand, for breast, lung, and gastrointestinal cancers, excellent agreement was observed between CODMR and CODDC in HL patients (corresponding Cohen’s kappa: 0.91, 0.79, and 0.83, respectively). HL patients more often died from cardiovascular disease according to CODMR than according to CODDC (20.9% vs. 15.6%, respectively). For ischemic heart disease, the conditional agreement was 57.8% for CODMR and 46.4% for CODDC, with a corresponding kappa value of 0.48. The conditional agreement for other heart diseases was low for CODMR (18.1%), and the corresponding kappa value of 0.24 reflected poor level of agreement between CODMR and CODDC. The overall agreement between CODMR and CODDC for common causes of death in HL patients was 81.4%, excluding other, unknown, and unavailable causes of death (Table 3). Including all other causes of death, the overall agreement was 70.8%. Among patients for whom the CODDC was HL, 29 out of 365 patients died from non-HL or leukemia according to CODMR. Similarly, 10 out of 180 patients who died of HL according to CODMR, died from non-HL or leukemia according to CODDC. Additionally, a substantial part of HL patients who died of HL according to CODDC died from circulatory disease according to CODMR (22 out of 164 patients).

Table 2 Level of agreement between causes of death from medical records and death certificates in HL patients

Notes: Cause of death according to the ICD-10: malignant neoplasms (C00–C97), gastrointestinal cancer (C15–C26), lung cancer (C33–C34), BC (C50), HL (C81), circulatory disease (I00–I99), ischemic heart disease (I20–I25), and other heart diseases (I30–I51). aConditional agreement in medical records is defined as number of deaths in agreement with death certificates divided by the total number of deaths from a specific cause of death. bConditional agreement in death certificates is defined as number of deaths in agreement with medical records divided by the total number of deaths from a specific cause of death.

Abbreviations: HL, Hodgkin lymphoma; BC, breast cancer.

Table 3 Causes of death according to medical records and death certificates in HL patients

Notes: Cause of death according to the ICD-10: HL (C81), BC (C50), lung cancer (C33–C34), gastrointestinal cancer (C15–C26), NHL and leukemia (C82–C96), circulatory disease (I00–I99), and respiratory disease (J00–J99). aUnknown and unavailable causes of death include ICD-10 codes R96–R99 and cases where no information on cause of death in medical records or no response from the GP could be obtained.

Abbreviations: HL, Hodgkin lymphoma; BC, breast cancer; NHL, non-Hodgkin lymphoma; GP, general practitioner.

In addition in BC patients, malignant neoplasm was the most important cause of death (Table 4). When the CODMR was BC, in 95.4% of deaths, the CODDC was also BC, resulting in a conditional agreement in CODDC of 95.0%. The corresponding Cohen’s kappa value of 0.75 indicated high agreement between CODMR and CODDC. Kappa values also indicated excellent agreement between CODMR and CODDC for lung cancer (0.78) and gastrointestinal cancer (0.80) in BC patients. For BC patients, a higher overall agreement between CODMR and CODDC for common causes of death was observed compared to HL patients (Table S1). Here, the overall agreement for breast, lung, and gastrointestinal cancers; other malignant neoplasms; and circulatory disease was 96.6%. Including other causes of death, the overall agreement was 91.7%.

Table 4 Level of agreement between causes of death from medical records and death certificates in BC patients

Notes: Cause of death according to the ICD-10: malignant neoplasms (C00–C97), gastrointestinal cancer (C15–C26), lung cancer (C33–C34), BC (C50), and circulatory disease (I00–I99). aConditional agreement in medical records is defined as number of deaths in agreement with death certificates divided by the total number of deaths from a specific cause of death. bConditional agreement in death certificates is defined as number of deaths in agreement with medical records divided by the total number of deaths from a specific cause of death.

Abbreviation: BC, breast cancer.

For 70 out of 109 deceased HL patients for whom CODMR and CODDC differed, death certificates were filed in the years that were available for review (Table 5). When CODMR was attributed to HL, seven discrepancies were due to the lack of specification by the physician, eg, NHL or “lymphoma” was stated on the death certificate rather than HL, and three discrepancies reflected coding errors by the registration clerk. When CODDC was recorded as HL, 16 discrepancies were caused by the fact that the physician filing the death certificate related the cause of death to late adverse events of treatment for HL, which was, however, diagnosed >10 years prior to death in patients cured of HL, eg, “lung fibrosis after radiotherapy for the treatment of HL” was stated on the death certificate.

Table 5 Causes of discrepancies between causes of death from medical records and causes of death from death certificates in HL patients

Notes: aThe death certificate stated that the patient committed suicide because there were no further options for the treatment of HL. bThe death certificate stated that the patient had euthanasia for BC. cNumbers do not add up to 100%; multiple comments possible when no explanation was found. dThe physician stated on the death certificate: “chemotherapy or radiotherapy for the treatment of HL”. eThe death certificate stated for example “cardiomyopathy after chemotherapy and radiotherapy for the treatment of HL”. fThe death certificate stated for example “heart failure”, whereas the medical records stated “cardiomyopathy”.

Abbreviations: HL, Hodgkin lymphoma; NHL, non-Hodgkin lymphoma; BC, breast cancer.

For 41 out of 42 deceased BC patients for whom we observed discrepancies between CODMR and CODDC, death certificates were available for review (Table S2). Of the 21 deaths attributed to BC in the CODMR but not in the CODDC, the discrepancy was once due to a coding error and the CODDC was once attributed to an external cause of death without BC being mentioned on the death certificate. No explanation could be found for the remaining 19 discrepancies. In none of the 20 discrepancies where CODDC was attributed to BC while the CODMR was recorded as another disease, an explanation could be found.

Table 6 shows the results of the person-year analysis of cause-specific mortality for common causes of death in HL patients, using CODMR complemented with CODDC when CODMR was missing and, alternatively, using CODDC only. Compared to SMRs based on CODDC only, SMRs based on CODMR complemented with CODDC were higher for circulatory disease (SMR=5.9; 95% CI=5.2–6.6 vs. SMR=4.6; 95% CI=4.0–5.2, P<0.05). Conversely, SMRs were lower for HL (SMR=1,030; 95% CI=917–1,152 vs. SMR=1,249; 95% CI=1,124–1,384, P<0.05). These differences are reflected in the AERs as well, especially for HL as cause of death. The SMRs in the person-year analysis, where we complemented CODMR with CODDC and where we supplemented CODMR with CODDC when these causes of death showed discrepancy, approximated the results of the analysis where we used CODDC only.

Table 6 Person-year analysis on cause-specific mortality using different cause-of-death data in HL patients

Notes: aCause of death was based on information from the medical records and supplemented with information from death certificates when the cause of death was unknown or unavailable from the medical records. bCause of death was based on information from the death certificates only. cCause of death was based on information from the medical records and supplemented with information from death certificates when the cause of death was unknown or unavailable from the medical records or when the cause of death was discrepant between the medical records and death certificates. dStatistical comparison between SMRs from the analysis using observed causes of death from medical records complemented with death certificates and using observed causes of death from death certificates only showed a statistically significant difference at a P-value of <0.05. Here, we assumed that the covariance between the different SMRs was zero. Since the two samples were not independent, covariance will be positive, resulting in a smaller P-value. Comparisons remain inconclusive when we observe a nonstatistically significant difference between SMRs at a covariance of zero.

Abbreviations: HL, Hodgkin lymphoma; NHL, non-Hodgkin lymphoma; O, observed; E, expected; SMR, standardized mortality ratio; AER, absolute excess risk.

Discussion

Overall, we observed high agreement when comparing causes of death from medical records and death certificates. Among HL patients, an overall agreement of 81% was observed for common causes of death. However, CODMR more often stated cardiovascular disease, whereas CODDC was more often recorded as HL. Differences between CODMR and CODDC were much smaller for BC patients. The majority of BC patients died from BC itself, and a high overall agreement (97%) was observed between CODMR and CODDC. We compared SMRs based on different sources of data in our HL cohort and observed a lower SMR for HL and a higher SMR for circulatory disease when using CODMR complemented with CODDC compared to using CODDC only. We only calculated SMRs for the HL cohort, since very few discrepancies were observed for common causes of death in the BC cohort, and only a small number of missing causes of death from the medical records would need to be complemented with the causes of death from the death certificate. Therefore, we do not expect to find more than minor differences in the calculated SMRs in the BC cohort. We assessed discrepancies at an individual level by reviewing original death certificates when differences were observed between CODMR and CODDC. However, no explanations could be found for most of these discrepancies. In our study, linkage errors are minimal and we do not expect that these errors explain the observed discrepancies. Part of the discrepancies between CODMR and CODDC in HL patients could be explained by the fact that the physician attributed the cause of death on the death certificate to a potential late adverse effect of treatment for HL. However, as the patient was cured of HL according to the medical record, recording of HL as a cause of death does not seem to be correct. In addition, for an individual patient, causality of an exposure-disease sequence can never be established. Here, we must keep in mind that CODMR and CODDC are collected for different purposes. In research focusing on the prognosis of HL or long-term adverse events of HL treatment, it is undesirable to code HL as the underlying cause of death when a patient is cured of HL and dies from a potential late effect of HL treatment. For example, when a physician states “cardiomyopathy as a consequence of chemotherapy for the treatment of HL”, the CODDC will be coded as HL, whereas in late effects research, the CODMR would be attributed to cardiomyopathy.

When interpreting the results of this study, it is important to keep in mind that CODMR and CODDC have been assessed in different ways. To obtain causes of death from HL and BC patients, we reviewed the medical files. If the cause of death could not be retrieved, a questionnaire was sent to the patient’s GP. Thus, where Statistics Netherlands has to rely on the physician who fills in the death certificate, in our research setting, we were able to utilize more information from medical records to obtain CODMR. Acknowledging these differences between CODMR and CODDC, the question arises what level of agreement can be reached when comparing CODMR and CODDC. Statistics Netherlands previously assessed the reliability of coding an underlying cause of death from death certificates.16 The reliability of the cause-of-death statistics, assessed by coding the same death certificate by different registration clerks, ranged from high (>90%) for major causes of death such as cancers and acute myocardial infarction to low (<70%) for chronic diseases. In the current study, we observed an overall agreement for common causes of death of 71% in HL patients and 92% in BC patients. Relating these findings to the reliability study of Statistics Netherlands, the level of agreement in the current study is comparable with known variability in determining cause of death.

Among HL patients, our most important finding was that CODMR was more often attributed to cardiovascular disease, whereas CODDC was more often attributed to HL. This indicates that HL is more often recorded in the cause-of-death statistics files than we would expect from a clinical perspective, leading to an overestimation of mortality from HL. Our comparison between CODMR and CODDC showed that discrepancies were more frequent within the HL cohort than those in the BC cohort. BC patients most often died from BC itself within a short period of time, whereas a substantial number of HL patients died years after the initial HL diagnosis. Although there are differences between HL and BC patients in the calendar period of cancer diagnosis, follow-up time, and age at treatment, discrepancies in cause of death appear to be more frequent when deaths occurred after a relatively long period of follow-up, when the likelihood of deaths due to late adverse events related to treatment is higher.

Conclusion

We observed high levels of agreement between CODMR and CODDC for common causes of death in HL and BC patients. However, differences were observed in the attribution of HL and circulatory disease as causes of death between CODMR and CODDC. Observed discrepancies between CODMR and CODDC frequently occurred in the presence of late effects of treatment for HL; HL was more often recorded as CODDC, whereas the CODMR more often concerned potential late effects of treatment. In future studies, researchers should keep in mind that differences in causes of death may arise when examining long-term mortality in cancer patients using different sources of cause-of-death data.

Disclosure

The authors report no conflicts of interest in this work.

References

1.

Johansson LA, Westerling R. Comparing Swedish hospital discharge records with death certificates: implications for mortality statistics. Int J Epidemiol. 2000;29(3):495–502.

2.

Johansson LA, Westerling R. Comparing hospital discharge records with death certificates: can the differences be explained? J Epidemiol Community Health. 2002;56(4):301–308.

3.

Johansson LA, Björkenstam C, Westerling R. Unexplained differences between hospital and mortality data indicated mistakes in death certification: an investigation of 1,094 deaths in Sweden during 1995. J Clin Epidemiol. 2009;62(11):1202–1209.

4.

Klijs B, Nusselder WJ, Mackenbach JP. Nationwide individual record linkage study showed poor agreement of causes of death and hospital diagnoses at individual level but reasonable agreement at population level. J Clin Epidemiol. 2014;67(2):160–168.

5.

Schaffar R, Rapiti E, Rachet B, Woods L. Accuracy of cause of death data routinely recorded in a population-based cancer registry: impact on cause-specific survival and validation using the Geneva Cancer Registry. BMC Cancer. 2013;13:609.

6.

Aleman BM, van den Belt-Dusebout AW, Klokman WJ, van’t Veer MB, Bartelink H, van Leeuwen FE. Long-term cause-specific mortality of patients treated for Hodgkin’s disease. J Clin Oncol. 2003;21(18):3431–3439.

7.

Ng AK, Bernardo MP, Weller E, et al. Long-term survival and competing causes of death in patients with early-stage Hodgkin’s disease treated at age 50 or younger. J Clin Oncol. 2002;20(8):2101–2108.

8.

Hooning MJ, Aleman BM, van Rosmalen AJ, Kuenen MA, Klijn JG, van Leeuwen FE. Cause-specific mortality in long-term survivors of breast cancer: a 25-year follow-up study. International journal of radiation oncology, biology. Physics. 2006;64(4):1081–1091.

9.

van Leeuwen FE, Klokman WJ, Hagenbeek A, et al. Second cancer risk following Hodgkin’s disease: a 20-year follow-up study. J Clin Oncol. 1994;12(2):312–325.

10.

van Leeuwen FE, Klokman WJ, Veer MB, et al. Long-term risk of second malignancy in survivors of Hodgkin’s disease treated during adolescence or young adulthood. J Clin Oncol. 2000;18(3):487–497.

11.

de Bruin ML, Burgers JA, Baas P, et al. Malignant mesothelioma after radiation treatment for Hodgkin lymphoma. Blood. 2009;113(16):3679–3681.

12.

Aleman BM, van den Belt-Dusebout AW, de Bruin ML, et al. Late cardiotoxicity after treatment for Hodgkin lymphoma. Blood. 2007;109(5):1878–1886.

13.

van Nimwegen FA, Schaapveld M, Janus CP, et al. Cardiovascular disease after Hodgkin lymphoma treatment: 40-year disease risk. JAMA Intern Med. 2015;175(6):1007–1017.

14.

Boekel NB, Jacobse JN, Schaapveld M, et al. Cardiovascular disease incidence associated with internal mammary chain irradiation and other breast cancer treatments: a large cohort study of patients treated from 1970 through 2009. Br J Cancer. Epub 2018 Aug 1.

15.

Organization WH. ICD-10: International Statistical Classification of Diseases and Related Health Problems. Geneva: WHO; 1992–1994.

16.

Harteloh P, de Bruin K, Kardaun J. The reliability of cause-of-death coding in The Netherlands. Eur J Epidemiol. 2010;25(8):531–538.

17.

van Sonsbeek JLA. Van de schaduw des doods tot een licht ten leven. De historie van de methodiek van de doodsoorzakenstatistiek in Nederland 1865-2005 [From the shadow of death to a light of life. The history of the methodology of cause-of-death statistics in the Netherlands 1865-2005]. Voorburg/Heerlen: Centraal Bureau voor de Statistiek; 2005. Dutch.

18.

Landis JR, Koch GG. The measurement of observer agreement for categorical data. Biometrics. 1977;33(1):159–174.

Supplementary materials

Table S1 Causes of death according to medical records and death certificates in BC patients

Notes: Cause of death according to the ICD-10: BC (C50), lung cancer (C33–C34), gastrointestinal cancer (C15–C26), and circulatory disease (I00–I99). aUnknown and unavailable causes of death include ICD-10 codes R96–R99 and cases where no information on cause of death in medical records or no response from the GP could be obtained.

Abbreviations: BC, breast cancer; GP, general practitioner.

Table S2 Causes of discrepancies between causes of death from medical records and causes of death from death certificates in BC patients

Note: aNumbers do not add up to 100%; multiple comments possible when no explanation was found.

Abbreviation: BC, breast cancer.

Creative Commons License © 2018 The Author(s). This work is published and licensed by Dove Medical Press Limited. The full terms of this license are available at https://www.dovepress.com/terms.php and incorporate the Creative Commons Attribution - Non Commercial (unported, v3.0) License. By accessing the work you hereby accept the Terms. Non-commercial uses of the work are permitted without any further permission from Dove Medical Press Limited, provided the work is properly attributed. For permission for commercial use of this work, please see paragraphs 4.2 and 5 of our Terms.