Analysis of Competing Risks of Causes of Death and their Variation Over Different Time Periods in Hodgkin's Disease

Hodgkin's disease is usually put forward as an example of a curable disease, but long-term studies show a greater risk of death among the population cured of lymphoma than among the general population. In an earlier study, we found an excess of mortality that was not due to the progression of the lymphoma when compared with the general population of the same age and gender. In this study, the standardized mortality ratio for patients with Hodgkin's disease was 6.8 [95% confidence intervals (95% CI), 5.4-8.4; P < 0.0001]; and the excess of death per 100 person-years in the overall sample of patients with Hodgkin's disease was 1.9. The standardized mortality ratio and excess of deaths were consistently >1 in all periods, even 20 years after diagnosis (1).

The slow clinical evolution of Hodgkin's disease makes the specific analysis of causes of death difficult and requires long follow-ups (2). We believe that, to avoid bias in studies of this kind, there has to be continuity and experience in the treatment team, and a prolonged and protocolized follow-up that guarantees a system of work in a population not previously selected on clinical trial criteria, which could give a false image of the results. In addition, studies of the causes of death usually evaluate the moment of the treatment of the patient: in the case of Hodgkin's disease, since the 1970s when curative treatment began until now, many changes have occurred, with the introduction of new drugs, methods of follow-up, histologic diagnostic criteria, etc. All these make it important to investigate the causes of death, and specifically, the changes during these 30 years.

The causes of death were studied through the model of competing risks, which provides a correct calculation of risk when many different competing risks of death exist.

A study was made of the clinical histories of 534 patients diagnosed with Hodgkin's disease at our hospital between January 1967 and September 2003. To be considered for inclusion, all patients had to have had a pathologically confirmed diagnosis of Hodgkin's disease. In the 56 cases in which the diagnosis was uncertain, the pathologic findings were reviewed and 21 cases in which the diagnosis remained unclear were excluded. For inclusion in the study, patients had to comply with the following criteria:

1. Extension and clinical classification study. All patients had to have an adequate extension study that covered their period of treatment and allowed their re-evaluation and stage classification according to the Rye meeting criteria, subsequently modified by Ann Arbor (3) and the Cotswolds consensus (4).

2. Treatments. The first chemotherapy treatment was the MOPP combination, as described by De Vita and Serpick (5), and consisting of the association of nitrogen mustard, vincristine, procarbazine, and prednisone. This was used in our Health Service until 1980. Since then, the ABVD guidelines designed by the Bonadonna group (6) have been used. ABVD is Adriamycin, bleomycin, vinblastine, and dacarbazine, either in isolation or alternating with MOPP. Radiotherapy treatment was given in accordance with the Kaplan (7) guidelines and followed in our hospital (8).

3. Survival time identification and definition. Overall survival: this is the entire time between diagnosis and the date of death or the last known date of the patient's remaining alive (9). All eligible patients and all deaths from whatever cause were included and the results expressed in months. Disease-free survival: this is the time between the end of treatment until the date of relapse or the last known date of remaining alive and being disease-free, or death only in the case of patients with complete remission (9). Patients were prospectively followed-up from their diagnosis every 3 months during the first 2 years, every 6 months between 2 and 5 years, and annually from the 5th year.

Study of time periods and causes of death. We studied the causes of death in three time periods: from diagnosis up to the introduction of anthracyclines with ABVD-type chemotherapy at our hospital in 1980 (called cohort A); then upon trying to establish groups with homogeneous patients between 1980 and 1985 (cohort B); and from then until the end of treatment in 2003, but with follow-up until December 31, 2006 (cohort C). Patients were allocated to each period on the basis of the date of diagnosis of the disease. We grouped events into five groups: (a) death due to the disease, (b) death due to second primary tumor, (c) death due to infectious disease, (d) cardiac, and (e) others.

Statistical analysis. The cumulative probability of survival was determined by the Kaplan-Meier method, whereas statistically significant differences between curves were evaluated with the log-rank test. The variability in overall survival among the different periods diagnostic was checked applying the Cox proportional regression model. The Mantel-Haenzel and χ² tests were used to compare categorical variables between groups. We studied the standardized mortality ratio, which is the ratio between the number of deaths observed in the cohort and the number of cases expected if the cohort had the same mortality rate as the general population. These data were available from, and was provided by, the Instituto Nacional de Estadística (National Statistics Institute, Madrid, Spain).

We analyzed the competing risks of secondary tumors and death from other causes, using cumulative incidence functions to calculate the fraction of patients suffering from each event over time (10). When multiple potential events occur, methodologies using cumulative incidence functions were used. The cumulative incidence of each individual event was calculated by the competing risks method, in which death from other causes was considered a competing risk. Sensitivity analysis calculated a maximum of 10 years follow-up of subjects, so that periods >10 years were artificially censored at 10 years.

For comparisons, Gray's test was used for the subdistribution between treatment groups and cause-specific mortality across calendar periods (11). Calculation, testing, and regression modeling with the hazards ratio and the corresponding 95% of subdistribution functions in competing risks for each cause in the period groups were achieved by Gray's test and the Fine and Gray method (12). Statistical analyses were done using the R software.⁵

We studied 534 patients (whose characteristics are listed in Table 1) with a median follow-up time of 9.1 years for all three cohorts. Median follow-up was 16 years for cohort A, 10.1 years for cohort B, and 7 years for cohort C. The 5-year, 15-year, and 20-year Kaplan-Meier survival estimates for all patients were 81%, 72%, and 65%, respectively. At the close of the study, 337 patients (63.1%) were alive, 170 (31.8%) had died, and 27 (5.1%) were lost.

Causes of death and overall risks of death. Fifty-three (31.1%) patients died of a progressed tumor. Most of these occurred in the 5 years following diagnosis, whereas 20% of the patients died between 5 and 15 years after diagnosis. After 15 years, there were no further deaths from tumor progression.

Seventy-six patients developed a second tumor and 48 patients died because of this, which made it the second leading cause of death with 28% of total exitus. The highest percentage of deaths from a second tumor occurred between 5 and 10 years from diagnosis (average 9.8 years), although for as long as 28 years, there continued to be deaths due to a second tumor.

The cumulative incidence of second tumors was 4.1% (95% CI, 2.7-6.2) at 5 years, 8.3% (95% CI, 6.1-11.2) at 10 years, and 22% (95% CI, 17-28.6) at 20 years from diagnosis. Secondary solid tumors (53 patients), including lung cancer (18), sarcomas (10), and breast cancer (19) were more common than secondary leukemia (13).

The third most important cause of death was infection (17%), with a median occurrence of 2.8 years after the end of treatment, although for up to 21 years, mortal infections continued to occur. Twenty-six patients (15% of the total) died because of the toxicity of the treatment.

Cardiovascular death occurred in nine patients (5.3% of the total). This is the cause of death with the latest average occurrence, 13.2 years median after the end of treatment, occurring from a few months to 27 years later. Table 2 breaks down the risks of death with 95% CIs by years since diagnosis and by cause of death.

Study of survival and causes of death by time periods from diagnosis. From 1967 to 1980, we treated 153 patients; from 1981 to 1985, 177 patients were treated; and since then, 204 patients have been treated. Figure 1 gives the likelihood of survival according to time periods. Overall survival from the period before 1980 was 88.6% at 5 years, 66.7% at 10 years, and 54.8% at 15 years, significantly worse (P = 0.001) than the other two cohorts. Cohort C had survival rates of 88%, 77%, and 73.7%, respectively. There were no statistically significant differences between the periods studied since 1981. Table 3 describes the clinical characteristics of each of the time periods indicated.

We studied the individual clinical characteristics of each cohort and compared them against each other. We found statistically significant differences between cohort B (1980-1985 period) and cohort A (<1980). In the former, there was a more nodular sclerosing histology to the detriment of lymphocyte depletion and mixed cellularity (P < 0.001), fewer advanced stages, fewer patients with B symptoms, and fewer combined treatments (all P < 0.001).

We found a similar pattern with the same statistically significant differences between cohorts A and C. However, the study of cohorts B and C found no differences between stages, treatments, frequency of B symptoms, or relapse. Although there continued to be differences in the histologic subtypes, mixed cellularity and lymphocyte depletion continued to decrease in their frequency of occurrence in favor of nodular sclerosing (P = 0.006).

The standardized mortality ratio was consistently high for all calendar periods under study, but they tended to diminish. The standardized mortality ratio was 13.9 for patients in cohort A (95% CI, 11.2-16.6; P < 0.0001), 7.1 in cohort B (95% CI, 6.4-7.9; P < 0.0001), and 5.5 in cohort C (95% CI, 8.2-4.6; P < 0.0001).

Table 4 describes the cumulative incidences and hazard ratios of causes of death at 5 and 10 years (analysis by calendar period). Although the number of patients who died decreased, tumor progression continued to be the main cause of death, followed by secondary tumors. Causes due to the toxicity of treatment and infection clearly decreased over time. Figure 1 shows in graphic format the occurrence in time, according to cause of death and time periods, of each cause of death taken individually.

Despite the advances introduced in the treatment of Hodgkin's disease, over the past three decades, we found the rate of mortality of our patients with Hodgkin's disease to be >10-fold higher than in the general population, with more than three deaths per 100 person-years. The long-term results and competing risk of causes of death and their variation over different time periods, given in this article, complement our previous data, in which we found a higher mortality rate in every calendar period. In particular, with the standardized mortality ratio (1), earlier periods, similar to cohort A, had the highest rates compared with the general population.

The survival times of our cohorts are in every way comparable with other cohorts (13), with 72% overall survival at 10 years, including the figure for advanced illness, and a prolonged mean follow-up of more than 9 years for the three cohorts together.

The clinical features of our patients at diagnosis only varied significantly in the identification of histologic subtypes, with a steadily lower frequency of patients with lymphocyte depletion in favor of nodular sclerosing. This corroborates the results of other series, which is consistent with epidemiologic studies of Western populations, in which a steady reduction in depletion and mixed cellularity is seen (14). Better diagnostic techniques, separating Hodgkin's disease from other non–Hodgkin lymphomas, along with their effect on exogenous lymphoma-related causes, could be the explanation for this phenomenon.

Comparatively, our results show that since 1980, our patients have been diagnosed earlier, at less advanced stages, and with fewer B symptoms and combined treatments. We believe they have been diagnosed and treated better (15, 16), and receive less toxicity than before. As a result, although the mortality rate tends to diminish over time, it is still quite high. There are few publications that have analyzed this in nonselected populations (17–21), and fewer still that analyze in detail the variation in causes of death over the years.

The main cause of death continues to be (in all years) the progression of disease, whether considered overall or individually by diagnostic periods, and this is despite the various diagnostic and therapeutic methods developed in these years. Although the success of the treatment is beyond doubt, there is still a population of incurable patients who continue to be an important subgroup. This group of patients who are refractory to treatment and die due to the progression of the tumor stands at ∼10% of all patients treated.

The hazards ratio for death from a second tumor suggests a reduction in the risk of death from a second tumor in patients treated prior to 1980, compared with the rest for whom there were no significant differences. A better understanding of the causes of death, fewer leukemia treatments such as MOPP (22), shorter treatments, and fewer combined treatments probably contributed to this improvement. In addition, screening programs, breast and skin self-examination, as well as oncologic treatments more suitable for these secondary tumors have become more common and could have modified the incidence of death due to a second tumor in this patient population, as the follow-up time in this population is also long. Treatments have become much more personalized, with fewer combined treatments, which dropped from 44.4% initially to 24%.

Change in infection as a cause of death is clearer. Thus, the hazards ratio of dying due to infection prior to 1980 was 3.5 times greater, with a clear statistical significance (P = 0.009). Cause of death from treatment-related toxicity behaves the same way: the hazards ratio was 7.1 times greater prior to 1980 (P = 0.001). Even so, the predisposition to developing serious infections may be linked to treatment procedures, such as laparatomy and splenectomy in the initial stage, done on patients systematically years ago, but now a practice in disuse. However, it is also known that before starting any treatment, patients present with disorders in their immune status. The blood lymphocytes of patients with untreated Hodgkin's disease are poorly activated by mitogens and antigens, which preferentially stimulate T cells (23). In addition, all patients with Hodgkin's disease should be advised of their increased risk of bacterial sepsis and opportunistic infections. In the Stanford series, the crude incidence of fatal infections among patients treated for Hodgkin's disease went down from 0.96% in the period before 1980 to 0.38% during the subsequent period (24).

Under all circumstances, causes of death, infections, and treatment toxicity were those on which prolonged follow-up had less influence and in which greater statistical variation was seen over time.

In addition, we ran the sensitivity analysis, which gives a maximum follow-up of 10 years, to try and avoid the bias that might be introduced by longer follow-up periods in some of the population groups analyzed; an alternative approach to summarizing cause-specific failure is based on quantiles using the cumulative incidence function (25). In any case, sensitivity analysis confirmed our results, i.e., it is not longer follow-up that determines the reduction in importance, or a magnification of it in case of second tumors, but rather a real variation in the causes of death (Table 4). Mortality for cardiac reasons occurred almost exclusively in those patients treated in cohort A, maybe due to insufficient follow-up in the other two cohorts because in our series, it is the cause of death that needs a greater period of follow-up, >13 years, to be identified. In these patients, there is a clear relationship between death and chest radiotherapy. Although cardiac deaths due to toxicity attributable to radiotherapy may appear many years later (26, 27), a reduction in these should be expected due to the use of new radiotherapy techniques. This could also explain why, since the 1980s, only two patients died due to this. This is in spite of the greater use of anthracyclines in the protocols. However, the experience of using ABVD indicates that myocardiopathy is rare in the absence of chest radiotherapy (28). Doses of anthracycline <450 to 500 mg/m² (29) are tolerated well without serious toxicity problems, unless there are predisposing factors present (30). It should also be remembered that the cumulative dosage in the treatment of Hodgkin's disease is lower, whether with MOPP-ABVD at ∼280 mg/m², or with six ABVD cycles at 350 mg/m².

In the light of our results, the question has to be raised as to whether the series on survival and causes of death that cover those patients under treatment since the 1970s fully reports the true situation. Compared with patients in the 1970s, there are now different causes of death which we have to deal with, which is important when we are planning preventive activities and clinical research. The progression of Hodgkin's disease is the main cause of death in all the periods studied. Over time, a clear reduction in death related to the toxicity of treatments was seen, although other deaths, such as those caused by second tumors or cardiovascular pathology (related to a great extent with radiotherapy treatments), continue to be a major cause of death among patients cured of Hodgkin's disease. Perhaps continuing along the line of avoiding combined treatments (31) might modify this trend, which has lasted without variation since the start of modern Hodgkin's disease treatment.