Hodgkin lymphoma (HL) survival in Sweden has improved dramatically over the last 40 years, but little is known about the extent to which efforts aimed at reducing long-term treatment-related mortality have contributed to the improved prognosis.

We used population-based data from Sweden to estimate the contribution of treatment-related mortality caused by diseases of the circulatory system (DCS) to temporal trends in excess HL mortality among 5,462 patients diagnosed at ages 19 to 80 between 1973 and 2006. Flexible parametric survival models were used to estimate excess mortality. In addition, we used recent advances in statistical methodology to estimate excess mortality in the presence of competing causes of death.

Excess DCS mortality within 20 years after diagnosis has decreased continually since the mid-1980s and is expected to further decrease among patients diagnosed in the modern era. Age at diagnosis and sex were important predictors for excess DCS mortality, with advanced age and male sex being associated with higher excess DCS mortality. However, when accounting for competing causes of death, we found that excess DCS mortality constitutes a relatively small proportion of the overall mortality among patients with HL in Sweden.

Excess DCS mortality is no longer a common source of mortality among Swedish patients with HL. The main causes of death among long-term survivors today are causes other than HL, although other (non-DCS) excess mortality also persists for as long as 20 years after diagnosis, particularly among older patients.

Survival after Hodgkin lymphoma (HL) has increased substantially over the last four decades, and for patients age younger than 65 years at diagnosis, the disease is now highly curable.1 The improved prognosis is likely attributable to improved patient assessment and staging, the development of effective multiagent chemotherapy, introduction of combined-modality therapy with reductions in radiation field size and dose, and more apt evaluation of treatment response. As a result of improvements in patient survival, research and clinical practice in recent decades have focused on understanding and reducing long-term treatment-related morbidity and mortality.27 In this article, we focus on excess mortality caused by cerebrovascular and cardiovascular diseases, with a particular focus on its absolute and relative contribution to the total excess mortality from HL.

Treatment-related mortality has typically been quantified using cause-specific mortality or excess mortality (ie, the difference between the observed and expected mortality rate in patients compared with a disease-free population). Both measures aim to provide estimates of the net survival associated with the disease (ie, survival in a hypothetical world where patients are assumed immune to death from causes other than the disease of interest).8 However, to accurately estimate the risk of, for example, death from treatment-induced cardiovascular disease, we should acknowledge that patients may also die from other causes (related or unrelated to HL) before they even become at risk for treatment-related adverse effects. Methods that account for competing risks have recently gained interest because they provide more useful measures of survival, compared with net survival, for risk communication purposes.9

We have recently developed statistical methodology that allows partitioning the excess mortality attributable to HL into component parts.10 Our method also provides estimates of crude probabilities of death (ie, probabilities estimated in the presence of competing risks). Although our focus is primarily on excess mortality as a result of diseases of the circulatory system (DCS), we also provide estimates of the remaining excess mortality as a result of other HL-related causes of death, including HL itself, second malignancies, infections, and so on. Application of this new methodology to population-based data from Sweden, with a homogeneous health care system and high-quality registers, provides additional insight into treatment-related DCS mortality. In particular, understanding the timing and magnitude of the crude probabilities of death from HL-related DCS is required to further develop optimal screening and prevention strategies for survivors of HL.

Data and Follow-Up Information

In Sweden, every physician and pathologist/cytologist is obliged by law to report each occurrence of cancer to the nationwide Swedish Cancer Registry.11 The registry was established in 1958 and contains basic information about all primary cancer diagnoses but not detailed clinical information about symptoms, disease stage (before 2002), and treatment. Deaths are recorded in the nationwide Causes of Death Register. A unique national registration number, assigned to all Swedish residents, enables matching of the registers.12

We identified all patients diagnosed with HL (International Classification of Diseases [ICD]-7:201) between 1973 and 2006 (N = 6,940). No patients were excluded based on previous cancer diagnoses, although in patients with multiple primary HL, we only considered the first diagnosis (n = 16 excluded). Patients younger than 19 years old at diagnosis (n = 597) were excluded because only six DCS deaths were observed among these patients. We further excluded patients older than age 80 years (n = 511) at diagnosis because the life expectancy is short compared with the expected time to develop treatment-induced DCS. Patients reported as having suspected/not microscopically verified HL or HL diagnosed incidentally at autopsy were excluded (n = 354), leaving 5,462 patients in the final cohort.

Among patients who died during follow-up, we classified the underlying cause of death as either DCS or other causes. DCS deaths were defined by codes 390 to 458 in the eighth revision of the ICD or its equivalent in other ICD revisions (Appendix, online only).13 We repeated this procedure in the entire Swedish Cause of Death Register and derived expected mortality rates for the two cause of death categories (ie, DCS and non-DCS), stratified by sex, age, and calendar year, as described in detail elsewhere.10 Patients were observed from the date of diagnosis until death, first emigration, or December 31, 2007, whichever occurred first, although maximum follow-up was restricted to 20 years. At the time the study was performed, cause of death was available to the end of 2007.

Statistical Methods

Excess mortality is defined as the difference between the all-cause mortality rate among the patients and the expected mortality rate in a comparable population, assumed free from the cancer of interest.14 Thus, excess mortality captures direct and indirect (eg, treatment-related) mortality, without using cause of death information. However, the main objective of this study is to identify two distinct components (DCS and non-DCS) that contribute to excess mortality from HL and to estimate their absolute and relative contribution to the total excess mortality. To achieve this, partitioning of the total excess mortality was done using flexible parametric survival models that enable modeling the two end points simultaneously.10 In this framework, the excess DCS mortality rate was obtained by comparing the observed DCS mortality rate among the patients (where all recorded DCS deaths count as events) with the expected DCS mortality rate in the entire Swedish population. Similarly, the remaining excess mortality rate was obtained by contrasting the mortality rate from deaths as a result of any other cause (non-DCS) among the patients to the expected non-DCS mortality rate in the background population. This approach assumes that individuals diagnosed with HL are comparable to the general Swedish population with respect to their risk of death from DCS and from other causes.

We estimated separate baseline excess mortality rate functions for the two distinct end points under study, which varied nonlinearly as a function of time since diagnosis using restricted cubic splines.15 All models were adjusted for sex, age at diagnosis, and calendar year of diagnosis (estimated using date of diagnosis in days). When investigating interaction terms between the variables, we sometimes kept nonsignificant effects in the model to avoid imposing constraints on estimates that were of primary interest. For example, we did not constrain the effect of calendar year to follow the same pattern for both the excess DCS mortality and the remaining excess mortality when studying temporal trends. In their simplest form, flexible parametric models assume proportional excess hazards. Under this assumption, the estimated excess mortality rate ratios are assumed constant throughout follow-up. We evaluated this assumption formally by including interaction terms between the variables under investigation and years since diagnosis and by assessing their statistical significance using likelihood ratio tests.

The results from modeling were combined with statistical methods for competing risks10,16 to predict past and future temporal trends in crude probabilities of death from the two sources of excess mortality. In these calculations, deaths from causes other than HL were introduced as a third possible outcome. A detailed description of the statistical models used for these predictions is provided in the Appendix.

The number of patients diagnosed was relatively constant over the study period (Table 1), although incidence rates have increased for patients up to age 35 and decreased for patients older than age 55 years at diagnosis.1,17 Almost half of the patients died within the follow-up period. Of these, 392 patients (14.8%) had DCS recorded as the underlying cause of death.


Table 1. Patient Demographic and Clinical Characteristics and Vital Status in the First 20 Years After Diagnosis Among Patients Diagnosed With Hodgkin Lymphoma in Sweden Between 1973 and 2006

Table 1. Patient Demographic and Clinical Characteristics and Vital Status in the First 20 Years After Diagnosis Among Patients Diagnosed With Hodgkin Lymphoma in Sweden Between 1973 and 2006

Characteristic All Patients (N = 5,462)
Deaths Caused by DCS
Deaths As Result of Other Causes
No. % No. % No. %
Year of diagnosis
    1973-1980 1,636 30.0 145 37.0 1,070 45.4
    1981-1988 1,252 22.9 139 35.5 593 26.2
    1989-1996 1,142 20.9 73 18.6 388 17.2
    1997-2006 1,432 26.2 35 8.9 209 9.2
Age at diagnosis, years
    19-35 1,896 34.7 20 5.1 281 12.4
    36-50 1,026 18.8 48 12.2 309 13.7
    51-65 1,151 21.1 116 29.6 645 28.5
    66-80 1,389 25.4 208 53.1 1,025 45.4
    Male 3,183 58.3 255 65.1 1,375 60.8
    Female 2,279 41.7 137 34.9 886 39.2
Total 5,462 100 392 14.8* 2,260 85.2*

Abbreviation: DCS, diseases of the circulatory system.

*Percentage of all deaths (any cause).

Table 2 lists the estimated excess DCS mortality rate ratios and the remaining excess mortality rate ratios from a proportional excess hazards model. Both the excess DCS mortality (P = .081) and the remaining excess mortality (P < .001) decreased over the study period. There was some evidence that the temporal trends in excess DCS mortality were statistically different from those of the remaining excess mortality (P = .10). For both end points, excess mortality was higher among males than females and among patients diagnosed at a more advanced age compared with younger patients.


Table 2. EMRRs and 95% CIs Partitioned Into Component Parts (ie, DCS and remaining excess mortality) From a Flexible Parametric Survival Model That Assumes Proportional Excess Hazards for All Covariate Effects

Table 2. EMRRs and 95% CIs Partitioned Into Component Parts (ie, DCS and remaining excess mortality) From a Flexible Parametric Survival Model That Assumes Proportional Excess Hazards for All Covariate Effects

Variable EMRR (DCS)
EMRR (remaining)
Ratio 95% CI Ratio 95% CI
Year of diagnosis
    1973-1980 1.00 Reference 1.00 Reference
    1981-1988 1.06 0.64 to 1.73 0.61 0.54 to 0.68
    1989-1996 0.54 0.27 to 1.10 0.42 0.37 to 0.48
    1997-2006 0.56 0.27 to 1.20 0.24 0.20 to 0.29
    P* .081 < .001
Age at diagnosis, years
    19-35 0.11 0.05 to 0.21 0.22 0.19 to 0.26
    36-50 0.42 0.24 to 0.74 0.42 0.36 to 0.49
    51-65 1.00 Reference 1.00 Reference
    66-80 2.26 1.28 to 4.00 2.03 1.81 to 2.27
    P* < .001 < .001
    Male 1.00 Reference 1.00 Reference
    Female 0.60 0.38 to 0.93 0.84 0.76 to 0.92
    P* .020 < .001

Abbreviations: DCS, diseases of the circulatory system; EMRR, excess mortality rate ratio.

*From a likelihood ratio test of heterogeneity.

Figure 1 shows the timing of the component-specific excess mortality rates (per 10,000 person-years) for two age groups of male patients diagnosed between 1973 and 1980 from a second model in which the proportional excess hazards assumption had been relaxed to reflect a more realistic scenario. The excess DCS mortality rate increases with years since diagnosis, although a high initial excess mortality is observed for patients between 51 and 65 years old at diagnosis. For the remaining excess mortality rate, the pattern is reversed.

Figure 2 shows temporal trends in the point estimates of the 5-, 10-, and 20-year excess mortality rates for two age groups of male patients estimated from a third model where year of diagnosis was modeled continuously and nonlinearly. In this model, nonproportional effects of year of diagnosis and age at diagnosis were estimated independently for the two end points, whereas the effect of sex was assumed to be proportional and shared. The excess DCS mortality remained relatively constant throughout the study period with some tendency toward a decrease in the short-term excess mortality from the mid-1980s.

For women, both the excess DCS mortality rates and the remaining excess mortality rates were lower compared with male patients (data not shown). For example, among young female patients, the difference in the 10-year excess mortality rates ranged from 1.5 to 1.0 per 10,000 person-years (for patients diagnosed in 1973 and 1993, respectively), and for older female patients, it ranged from 12.8 to 8.9 per 10,000 person-years.

Figure 3 shows the estimated crude probabilities of death from different causes for men diagnosed in 1987 and 2000 at ages 30 and 60, respectively. Irrespective of date of diagnosis, excess DCS deaths constitute a relatively small proportion of the total probability of death. For both young and old patients, deaths from other HL-related causes remain the principal source of cancer-associated mortality throughout the first 20 years after diagnosis.

The past and predicted future temporal trends in the 20-year crude probabilities of death from different causes are summarized in Figure 4. For men age 30 years at diagnosis, the relative contribution of excess DCS deaths to the total probability of dying is comparable to that of deaths from other causes than cancer throughout the entire observable study period (Fig 4A). For patients age 60 years at diagnosis, the corresponding probabilities are substantially larger than for young patients (Fig 4C). For patients diagnosed before 1989, the long-term crude probabilities of dying from any other HL-related causes have decreased dramatically. The predicted future crude probabilities of death suggest a continually decreased risk for patients diagnosed up until year 2003 (Figs 4B and 4D). The predicted future burden of excess DCS deaths remains small for young patients, whereas it is predicted to decrease for patients diagnosed at age 60. Deaths from other causes are predicted to further increase for older patients diagnosed between 1989 and 2003.

This study shows how the long-term excess DCS mortality experienced by patients with HL in Sweden, diagnosed and treated in the 1990s and early 2000s, is expected to further decrease compared with that of patients diagnosed in earlier years. A similar, but stronger, trend is also observed for the remaining excess HL mortality. The improvements in survival are not likely to be explained by changes in disease characteristics (ie, a more a favorable distribution of prognostic factors in recent years).1 The highest excess DCS mortality, both short and long term, was observed in patients diagnosed before the mid-1980s, reflecting the widespread use of mantle radiotherapy and total nodal irradiation. In addition, in patients with advanced (stage IIB to IV) and bulky disease, slow tumor regression, and/or evidence of residual disease, radiotherapy was given after combination chemotherapy. Since the start of the Stockholm Hodgkin lymphoma Study Group in 1973, treatment of HL in Sweden has been typical of treatment in other high-income countries. Mechlorethamine, vincristine, procarbazine, and prednisolone (MOPP); MOPP alternated with doxorubicin, bleomycin, vinblastine, and dacarbazine (ABVD); and ABVD alone were the most frequently used combination chemotherapies in advanced-stage HL during the study period, although alternative chemotherapy regimens have also been used, primarily among older patients and in patients with refractory or relapsing disease.1,18,19 Cumulative doses exceeding 550 mg/m2 of doxorubicin and 300 mg/m2 of bleomycin were discouraged already in 1985, because of the known toxicities of these drugs. In the late 1980s, combined-modality therapy, with successive reductions in radiation field size and dose, was introduced for patients with limited-stage disease.1,18,20,21

Our data show that excess DCS mortality has primarily been a concern among older patients. Pre-existing comorbid conditions have been reported in more than half of patients with HL older than age 60 years22 and provide a possible mechanism to produce treatment toxicity in elderly patients.4,9,23 We found no evidence suggesting that the probability of death from excess DCS has increased over time as a consequence of an increasing number of long-term survivors. However, the proportion of deaths from other causes than HL has increased continually, primarily among older patients. The latter types of death encompass all causes of deaths that can be expected in a population without HL (including cerebrovascular and cardiovascular deaths). Hence, although deaths from causes other than HL (including the expected deaths from DCS) are predicted to increase as a result of an increasing number of survivors, there is no such corresponding trend for the predicted excess DCS deaths. In the Swedish National Care Programme, which commenced in 1985 and is updated regularly, principles for staging and treatment were introduced, and reducing long-term adverse events (by reducing radiotherapy) was included as an important end point in patients with early- and intermediate-stage disease.1,20,21 Efforts to reduce treatment-related cardiotoxicity also included rigorous quit smoking advice, blood pressure management, and eventually echocardiography screening, and these efforts will contribute to the observed reduction in long-term excess DCS mortality. This program has been extended during recent years.24 However, reports from the Swedish National Board of Health and Welfare show that the 28-day lethality from acute myocardial infarction has decreased, for both men and women, since 1987, which is likely to contribute to our results.25 Moreover, the age-adjusted lethality among men is reported to be higher than for women, which can possibly explain why women also experience lower excess DCS mortality compared with men.

It has been speculated that the shift from extended-field radiotherapy toward involved-field radiotherapy among patients receiving mediastinal irradiation does not necessarily reduce the dose to the proximal coronary arteries, suggesting that the risk of coronary artery disease after radiotherapy treatment should remain a concern for patients managed in the modern era.26 However, Hull et al27 reported a significantly higher risk of coronary artery disease in patients treated with extended-field radiotherapy compared with patients treated with mantle fields. In addition, with new developments such as involved node and intensity-modulated radiotherapy, significant vascular-sparing effects are hopefully achieved.28,29

Although combined-modality treatment seems to have been successful in reducing toxicity without compromising the antitumor effects, by restricting low-dose radiotherapy and reducing chemotherapy cycles, the long-term effects (> 15 years) of modern HL treatment are still largely unknown.3032 Nevertheless, a growing literature shows that late effects are indeed significantly correlated with radiation dose and field size.33,34 Moreover, the results from a recent randomized comparison of patients with limited-stage HL treated with ABVD alone, single-modality subtotal nodal irradiation, or combined-modality ABVD plus subtotal nodal irradiation do show favorable 12-year overall survival in the group randomly assigned to therapy that excludes radiotherapy.35 The authors conclude that ABVD alone is associated with fewer deaths from causes other than the underlying disease, such as late effects from the treatment.

Information about initial treatment was lacking in our study, which precluded us from disentangling the relative contribution of anthracyclines and average radiation doses and fields on the excess DCS mortality. However, the results of this study can provide us with important clues about what the future holds for long-term survivors of HL. Most importantly, our results suggest that excess DCS mortality is not a common cause of death among Swedish patients with HL. However, we have not investigated DCS morbidity among these patients. Thus, we are not suggesting that DCS morbidity is no longer a concern for long-term survivors of HL in Sweden.

Potential sources for bias in this study include violations of the assumption that patients with HL are comparable to the general Swedish population. If, for example, the composition of patients with HL differs from the general population with respect to social class (eg, with young patients with HL having higher social class and, thereby, on average, better survival), the reported excess mortality rates would be too low.36 Moreover, the 20-year restriction of follow-up implies that, among patients with a long life expectancy, our conclusions related to the long-term prognosis may be too optimistic. Thus, our findings must be interpreted within the context of the limitation of observational data.

In conclusion, this population-based study from Sweden corroborates findings from several large studies that have investigated the risk of fatal cardiovascular disease among long-term survivors of HL. We use recent advances in statistical methodology to provide clinically interpretable estimates of temporal trends in the burden of fatal excess DCS toxicity facing HL survivors who were treated with regimens that are, to some extent, outdated in modern HL management. In addition, we used the modeling approach to predict the future clinical burden of excess DCS deaths among patients diagnosed and treated in the modern era of HL management. Although the panorama of possible late effects attributable to modern HL treatment has yet to be verified, we believe that the methods used in this study maximize the amount of information that can be obtained from the available data. Our findings provide evidence toward further reductions in both excess DCS mortality and remaining excess mortality, which indicate that long-term survivors in Sweden are doing quite well, although more can be done with respect to the non-DCS mortality from HL, particularly in elderly patients.

© 2013 by American Society of Clinical Oncology

Supported by Grants No. CAN 2010/676 (P.W.D.), CAN 2009/1012 (P.C.L.), and CAN 2009/1203 (M.B.) from the Swedish Cancer Society and by the Adolf H. Lundin Charitable Foundation (M.B.).

Presented in part at the 2012 North American Association of Central Cancer Registries Annual Conference, June 2-8, 2012, Portland, OR, and the 2011 Annual Meeting of the Association of Nordic Cancer Registries, August 31-September 2, 2011, Åland, Finland.

Authors' disclosures of potential conflicts of interest and author contributions are found at the end of this article.

The author(s) indicated no potential conflicts of interest.

Conception and design: Sandra Eloranta, Jan Sjöberg, Magnus Björkholm, Paul W. Dickman

Financial support: Paul W. Dickman

Collection and assembly of data: Sandra Eloranta

Data analysis and interpretation: All authors

Manuscript writing: All authors

Final approval of manuscript: All authors

1. J Sjöberg, C Halthur, SY Kristinsson , etal : Progress in Hodgkin lymphoma: A population-based study on patients diagnosed in Sweden 1973-2009 Blood 119: 990996,2012 Crossref, MedlineGoogle Scholar
2. BM Aleman, AW van den Belt-Dusebout, WJ Klokman , etal : Long-term cause-specific mortality of patients treated for Hodgkin's disease J Clin Oncol 21: 34313439,2003 LinkGoogle Scholar
3. BM Aleman, FE van Leeuwen : Are we improving the long-term burden of Hodgkin's lymphoma patients with modern treatment? Hematol Oncol Clin North Am 21: 961975,2007 Crossref, MedlineGoogle Scholar
4. S Myrehaug, M Pintilie, L Yun , etal : A population-based study of cardiac morbidity among Hodgkin lymphoma patients with preexisting heart disease Blood 116: 22372240,2010 Crossref, MedlineGoogle Scholar
5. AK Ng, MP Bernardo, E Weller , etal : 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 20: 21012108,2002 LinkGoogle Scholar
6. AJ Swerdlow, CD Higgins, P Smith , etal : Myocardial infarction mortality risk after treatment for Hodgkin disease: A collaborative British cohort study J Natl Cancer Inst 99: 206214,2007 Crossref, MedlineGoogle Scholar
7. RE van Rijswijk, J Verbeek, C Haanen , etal : Major complications and causes of death in patients treated for Hodgkin's disease J Clin Oncol 5: 16241633,1987 LinkGoogle Scholar
8. MP Perme, J Stare, J Estève : On estimation in relative survival Biometrics 68: 113120,2012 Crossref, MedlineGoogle Scholar
9. BM Aleman, AW van den Belt-Dusebout, ML De Bruin , etal : Late cardiotoxicity after treatment for Hodgkin lymphoma Blood 109: 18781886,2007 Crossref, MedlineGoogle Scholar
10. S Eloranta, PC Lambert, TM-L Andersson , etal : Partitioning of excess mortality in population-based cancer patient survival studies using flexible parametric survival models BMC Med Res Methodol 12: 86,2012 Crossref, MedlineGoogle Scholar
11. L Barlow, K Westergren, L Holmberg , etal : The completeness of the Swedish Cancer Register: A sample survey for year 1998 Acta Oncol 48: 2733,2009 Crossref, MedlineGoogle Scholar
12. JF Ludvigsson, P Otterblad-Olausson, BU Pettersson , etal : The Swedish personal identity number: Possibilities and pitfalls in healthcare and medical research Eur J Epidemiol 24: 659667,2009 Crossref, MedlineGoogle Scholar
13. IM Moriyama : The eighth revision of the International Classification of Diseases Am J Public Health Nations Health 56: 12771280,1966 Crossref, MedlineGoogle Scholar
14. PW Dickman, HO Adami : Interpreting trends in cancer patient survival J Intern Med 260: 103117,2006 Crossref, MedlineGoogle Scholar
15. S Durrleman, R Simon : Flexible regression models with cubic splines Stat Med 8: 551561,1989 Crossref, MedlineGoogle Scholar
16. PC Lambert, PW Dickman, CP Nelson , etal : Estimating the crude probability of death due to cancer and other causes using relative survival models Stat Med 29: 885895,2010 Crossref, MedlineGoogle Scholar
17. H Hjalgrim, J Askling, E Pukkala , etal : Incidence of Hodgkin's disease in Nordic countries Lancet 358: 297298,2001 Crossref, MedlineGoogle Scholar
18. M Björkholm, E Svedmyr, J Sjoberg : How we treat elderly patients with Hodgkin lymphoma Curr Opin Oncol 23: 421428,2011 Crossref, MedlineGoogle Scholar
19. GP Canellos, JR Anderson, KJ Propert , etal : Chemotherapy of advanced Hodgkin's disease with MOPP, ABVD, or MOPP alternating with ABVD N Engl J Med 327: 14781484,1992 Crossref, MedlineGoogle Scholar
20. B Glimelius, G Enblad, M Kalkner , etal : Treatment of Hodgkin's disease: The Swedish National Care Programme experience Leuk Lymphoma 21: 7178,1996 Crossref, MedlineGoogle Scholar
21. CG Raud, G Enblad, B Melin , etal : Evaluation of the Nordic Study for Early Stage Hodgkin Lymphoma Ann Oncol 22: 179180,2011 Google Scholar
22. DJ van Spronsen, ML Janssen-Heijnen, VE Lemmens , etal : Independent prognostic effect of co-morbidity in lymphoma patients: Results of the population-based Eindhoven Cancer Registry Eur J Cancer 41: 10511057,2005 Crossref, MedlineGoogle Scholar
23. C Glanzmann, P Kaufmann, R Jenni , etal : Cardiac risk after mediastinal irradiation for Hodgkin's disease Radiother Oncol 46: 5162,1998 Crossref, MedlineGoogle Scholar
24. A Andersson, U Näslund, B Tavelin , etal : Long-term risk of cardiovascular disease in Hodgkin lymphoma survivors: Retrospective cohort analyses and a concept for prospective intervention Int J Cancer 124: 19141917,2009 Crossref, MedlineGoogle Scholar
25. Myocardial infarctions in Sweden 1987-2010. Report series: Official statistics of Sweden. Statistics – Health and Medical Care 2425,2011 The National Board of Health and Welfare: Stockholm, Sweden The National Board of Health and Welfare Google Scholar
26. DC Hodgson : Late effects in the era of modern therapy for Hodgkin lymphoma Hematology Am Soc Hematol Educ Program 2011: 323329,2011 Crossref, MedlineGoogle Scholar
27. MC Hull, CG Morris, CJ Pepine , etal : Valvular dysfunction and carotid, subclavian, and coronary artery disease in survivors of Hodgkin lymphoma treated with radiation therapy JAMA 290: 28312837,2003 Crossref, MedlineGoogle Scholar
28. M Ghalibafian, A Beaudre, T Girinsky : Heart and coronary artery protection in patients with mediastinal Hodgkin lymphoma treated with intensity-modulated radiotherapy: Dose constraints to virtual volumes or to organs at risk? Radiother Oncol 87: 8288,2008 Crossref, MedlineGoogle Scholar
29. A Paumier, M Ghalibafian, A Beaudre , etal : Involved-node radiotherapy and modern radiation treatment techniques in patients with Hodgkin lymphoma Int J Radiat Oncol Biol Phys 80: 199205,2011 Crossref, MedlineGoogle Scholar
30. GP Canellos, D Niedzwiecki, JL Johnson : Long-term follow-up of survival in Hodgkin's lymphoma N Engl J Med 361: 23902391,2009 Crossref, MedlineGoogle Scholar
31. GP Canellos, JS Abramson, DC Fisher , etal : Treatment of favorable, limited-stage Hodgkin's lymphoma with chemotherapy without consolidation by radiation therapy J Clin Oncol 28: 16111615,2010 LinkGoogle Scholar
32. DL Longo : Radiation therapy in the treatment of Hodgkin's disease: Do you see what I see? J Natl Cancer Inst 95: 928929,2003 Crossref, MedlineGoogle Scholar
33. ML De Bruin, J Sparidans, MB van't Veer , etal : Breast cancer risk in female survivors of Hodgkin's lymphoma: Lower risk after smaller radiation volumes J Clin Oncol 27: 42394246,2009 LinkGoogle Scholar
34. S Sasse, B Klimm, H Görgen , etal : Comparing long-term toxicity and efficacy of combined modality treatment including extended- or involved-field radiotherapy in early-stage Hodgkin's lymphoma Ann Oncol 23: 29532959,2012 Crossref, MedlineGoogle Scholar
35. RM Meyer, MK Gospodarowicz, JM Connors , etal : ABVD alone versus radiation-based therapy in limited-stage Hodgkin's lymphoma N Engl J Med 366: 399408,2012 Crossref, MedlineGoogle Scholar
36. SL Glaser, RF Jarrett : The epidemiology of Hodgkin's disease Baillieres Clin Haematol 9: 401416,1996 Crossref, MedlineGoogle Scholar


We thank George P. Canellos, MD, for insightful comments on an early draft of this article.

List of Conditions Used to Define Deaths From Diseases of the Circulatory System

Following is a list of the conditions that were used to define deaths from diseases of the circulatory system (DCS) based on the eighth revision of the International Classifications of Diseases.13

(390-392) Acute rheumatic fever

390 Rheumatic fever without mention of heart involvement

391 Rheumatic fever with heart involvement

392 Chorea

(393-398) Chronic rheumatic heart disease

393 Diseases of pericardium

394 Diseases of mitral valve

395 Diseases of aortic valve

396 Diseases of mitral and aortic valves

397 Diseases of other endocardial structures

398 Other heart disease, specified as rheumatic

(400-404) Hypertensive disease

400 Malignant hypertension

401 Essential benign hypertension

402 Hypertensive heart disease

403 Hypertensive renal disease

404 Hypertensive heart and renal disease

(410-414) Ischemic heart disease

410 Acute myocardial infarction

411 Other acute and subacute forms of ischemic heart disease

412 Chronic ischemic heart disease

413 Angina pectoris

414 Asymptomatic ischemic heart disease

(420-429) Other forms of heart disease

420 Acute pericarditis, nonrheumatic

421 Acute and subacute endocarditis

422 Acute myocarditis

423 Chronic disease of pericardium, nonrheumatic

424 Chronic disease of endocardium

425 Cardiomyopathy

426 Pulmonary heart disease

427 Symptomatic heart disease

428 Other myocardial insufficiency

429 Ill-defined heart disease

(430-438) Cerebrovascular disease

430 Subarachnoid hemorrhage

431 Cerebral hemorrhage

432 Occlusion of precerebral arteries

433 Cerebral thrombosis

434 Cerebral embolism

435 Transient cerebral ischemia

436 Acute but ill-defined cerebrovascular disease

437 Generalized ischemic cerebrovascular disease

438 Other and ill-defined cerebrovascular disease

(440-448) Diseases of arteries, arterioles, and capillaries

440 Arteriosclerosis

441 Aortic aneurysm (nonsyphilitic)

442 Other aneurysm

443 Other peripheral vascular disease

444 Arterial embolism and thrombosis

445 Gangrene

446 Polyarteritis nodosa and allied conditions

447 Other diseases of arteries and arterioles

448 Diseases of capillaries

(450-458) Diseases of veins and lymphatics, and other diseases of circulatory system

450 Pulmonary embolism and infarction

451 Phlebitis and thrombophlebitis

452 Portal vein thrombosis

453 Other venous embolism and thrombosis

454 Varicose veins of lower extremities

455 Hemorrhoids

456 Varicose veins of other sites

457 Noninfective disease of lymphatic channels

458 Other diseases of circulatory system

Cerebrovascular and cardiovascular diseases were combined into one single end point to gain statistical power. Among cardiovascular deaths, myocardial infarction is the single most frequently occurring and investigated late effect from Hodgkin lymphoma (HL) treatment. Excess mortality from cerebrovascular diseases has not been studied to the same extent, although the body of evidence reporting on excess morbidity from this group of conditions, mainly stroke and transient ischemic attacks, is growing (De Bruin ML, Dorresteijn LD, van't Veer MB, et al: J Natl Cancer Inst 101:928-937, 2009; Dorresteijn LD, Kappelle AC, Boogerd W, et al: J Clin Oncol 20:282-288, 2002; Bowers DC, McNeil DE, Liu Y, et al: J Clin Oncol 23:6508-6515, 2005).

Statistical Methods

Here, we describes the statistical methods and model used to predict the future 20-year probabilities of death from various causes among patients diagnosed with HL in Sweden between the years 1973 to 2003 (Fig 4).

Model specification.

Flexible parametric survival models adapted for relative survival can be used to estimate the excess mortality rate attributable to a diagnosis of HL by subtracting the mortality rate that we would expect to observe in a comparable, disease-free group in the general population from the observed (all-cause) mortality rate among patients with HL.

The estimated excess mortality can be interpreted as the additional mortality rate that the patients experience, either directly (eg, because of refractory/relapsing disease) or indirectly (eg, after fatal late effects of treatment) as a result of the diagnosis of cancer.

To study the absolute and relative contribution of excess mortality attributable to diseases in the cardiovascular and cerebrovascular system, we further partitioned the observed all-cause mortality into smaller components in the following way:

Written in this form, it becomes evident that the excess DCS mortality is the additional DCS mortality that patients with HL experience on top of what we would expect them to have experienced had they never been diagnosed with HL. Similarly, we can refer to the second component part of the excess mortality as the remaining excess mortality attributable to the HL diagnosis.

To achieve this, we fitted a flexible parametric survival model that simultaneously estimated the excess DCS mortality rate and the remaining excess mortality rate attributable to a diagnosis of HL.10 We used a restricted cubic spline with 5 df (where the interior knots were placed at the 20th, 40th, 60th, and 80th centiles of the distribution of the uncensored log survival times, whereas the boundary knots were placed at the extremes of the same distribution) to model the two component-specific baseline excess hazard functions. Both end points were adjusted for sex, age at diagnosis, and calendar year of diagnosis. The effects of age and calendar year were modeled continuously and nonlinearly using restricted cubic splines, each with 4 df where the knots were placed at the fifth, 25th, 50th, 75th, and 95th percentiles of the distribution of the two variables. The effects of age at diagnosis and calendar year of diagnosis on the excess mortality rates were estimated independently and time dependently (ie, accounting for nonproportional excess hazards) when modeling the two end points. Each time-dependent effect was modeled with 12 df (3 df × 4 parameters representing the basis vectors for the age and calendar effects, respectively). Moreover, the structure/shape of the time-dependent effect of calendar year of diagnosis was assumed independent for the two outcomes by including a three-way interaction where 4 df were used to model the departure from a model with a shared time-dependent effect for the two outcomes. In all, 57 parameters were estimated in the final flexible parametric model. The model was estimated using the stpm2 module in the Stata software (StataCorp, College Station, TX).

Predicting future long-term excess mortality rates to estimate the 20-year crude probabilities of death.

Among all patients with HL included in the study, we were only able to fully observe and estimate the 20-year excess DCS mortality rate and remaining excess HL mortality rate for patients diagnosed before 1989 (because of administrative censoring on December 31, 2007). For patients diagnosed from 1989 and onward, only partial follow-up was observable. For example, we were able to observe patients diagnosed in the year 2000 for death or censoring for a maximum of 7 years. To further illustrate this, the predicted excess DCS mortality rates (per 10,000 person-years) for men age 60 years at diagnosis and diagnosed in the years 1980, 1985, 1990, 1995, and 2000, estimated within the range of observable data, are provided in Figure A1A.

To predict the future long-term excess mortality rates for the two component parts, we used our flexible parametric model to extrapolate predictions of the 20-year excess DCS mortality beyond the range of observable data. When predicting outside the range of observable data, the flexible parametric survival model assumes linearity (on the log cumulative excess hazards scale) beyond the last knot of the spline, which is used to represent the time scale in the analysis. The predicted future long-term excess DCS mortality rates are represented by the dashed lines in Figure A1B. Figure A2 shows the corresponding patterns for the remaining excess mortality rates for the same groups of patients.

Apart from any assumptions that were made in terms of specifying a model for the covariate effects, the extrapolated rates should only be interpreted under the additional assumption that no unpredictable changes in the general pattern and trends of the mortality of either patients with HL or the general population occur.

Estimating crude probabilities of death as a result of specific causes of death (Fig 4).

The estimated component-specific excess mortality rates were used to calculate crude probabilities of death using methodology previously described by Eloranta et al10 and Lambert et al.16 When calculating the crude probabilities of death outside the range of observable follow-up, we further assumed that the expected mortality rates (ie, expected DCS mortality rate and non-DCS mortality rate) in the Swedish background population (stratified on sex, age, and calendar year) would remain fixed at the 2007 level for all years from 2008 through 2023. Year 2023 was the last year for which background mortality rates were required to calculate the predicted future long-term crude probabilities of death for patients diagnosed between 1989 and 2003.


No companion articles


DOI: 10.1200/JCO.2012.45.2714 Journal of Clinical Oncology 31, no. 11 (April 10, 2013) 1435-1441.

Published online February 25, 2013.

PMID: 23439756

ASCO Career Center