¹Department of Internal Medicine, Section of Hematology/Oncology, University of Chicago, IL, USA

²Section of Cardiology, University of Chicago, IL, USA

Correspondence to: Dr B Brockstein, Assistant Professor of Medicine, Northwestern University Medical School Evanston Northwestern Healthcare, Evanston Hospital, Section of Hematology/Oncology, 2650 Ridge Ave, Evanston, IL 60201, USA

We sought to define risk factors predisposing breast cancer and lymphoma patients to cardiac and pulmonary toxicity when undergoing high-dose chemotherapy (HDC) and autologous stem cell rescue (ASCR). Additionally, we evaluated in depth the predictive value of the ejection fraction measured prior to HDC in determining cardiac toxicity. In this retrospective analysis, 24 variables were examined in 138 patients undergoing HDC and ASCR from 1990 until 1995. Logistic regression models were used to model the probability of experiencing cardiac and pulmonary toxicity as a function of the 24 prognostic covariates. Cardiac toxicity occurred in 12% of patients and pulmonary toxicity in 24% of patients. Bivariate analyses showed that patients with lymphoma (as opposed to breast cancer) and those with a higher cardiac risk factor score were more likely to experience cardiac toxicity. Multivariate logistic regression models predicted lymphoma and older age to be risk factors for cardiac toxicity. History of an abnormal ejection fraction and higher doses of anthracyclines prior to HDC may also contribute to cardiac toxicity. Pulmonary toxicity occurred more commonly in lymphoma than breast cancer patients, likely due to the busulfan used in the HDC regimen. No other risk factors for pulmonary toxicity were identified. We conclude that older patients with lymphoma should be carefully evaluated prior to being accepted for HDC programs. Older patients with breast cancer may tolerate this procedure well. There is a trend towards cardiac toxicity in patients with a past history of low ejection fraction, although seemingly poor cardiac risk patients may fare well with HDC if carefully selected with the aid of a thorough cardiac evaluation. Bone Marrow Transplantation (2000) 25, 885-894.

cardiac toxicity; pulmonary toxicity; stem cell transplant; ejection fraction; breast cancer; lymphoma

High-dose chemotherapy with autologous stem cell rescue is widely used to treat a variety of neoplastic diseases including breast cancer,^1,2 lymphoma,³ leukemia,⁴ and germ cell tumors.⁵ Improvements in conditioning regimens and advances in post-transplant supportive care have helped to decrease the toxicities and cost of this procedure. Unfortunately, non-hematologic toxicities including cardiac and pulmonary toxicity remain a problem. The risk factors for the development of these toxicities, however, are not clear. An understanding of these risk factors may help in the selection of appropriate patients for these procedures.

Drug-related cardiac toxicity in patients treated with high-dose chemotherapy has been well described. Cardiac toxicity due to cyclophosphamide was initially described as a complication of bone marrow transplantation by Santos et al⁶ and subsequently by others.^7,8,9 Several other chemotherapy drugs used in high-dose chemotherapy can be cardiotoxic. These include the anthracyclines,^10,11,12,13 and ifosfamide.^14,15 Other chemotherapy drugs given at non-transplant doses such as 5-fluorouracil,^16,17 paclitaxel,^18,19 vincristine,²⁰ and mitomycin C²¹ may induce cardiac toxicity. In addition to direct drug-related effects, other risk factors have been described for cardiac toxicity in patients who receive chemotherapy. Radiation therapy to the chest wall may directly cause cardiac toxicity or augment the effects of chemotherapy.^22,23,24 Potentially cardiotoxic effects of the transplant itself,²⁵ and the use of the preservative dimethyl sulfoxide (DMSO)²⁶ have also been reported. Risk factors other than these are less clear. Nonetheless, patients may be excluded from high-dose chemotherapy on the basis of suspected predisposing factors such as EKG abnormalities, or low pre-transplant ejection fraction.

Several previously published studies have aimed at identifying predictive factors which could lead to the development of cardiac toxicity. In one, it was found that there was no correlation between cardiac toxicity and the results of a cardiologic evaluation before bone marrow transplantation.²⁷ In another, a trend was found toward the development of cardiac toxicity in patients with a pretreatment ejection fraction of less than 50%.²⁸ These findings underscore the need for further study.

Pulmonary toxicity resulting from bone marrow transplantation has also been well documented. Carmustine (BCNU), busulfan and other drugs may cause pulmonary toxicity, and pulmonary infections are common.^{29,30,31,32,33} In patients undergoing allogeneic transplantation, rates of pulmonary toxicity as high as 40-60% have been described.^31,32 Although less common, pulmonary toxicity is still significant in patients receiving autologous bone marrow or stem cell transplants.³⁴ The factors which predict pulmonary toxicity have been more thoroughly studied than those predicting cardiac toxicity. These published predictive factors, however, are not consistent. For example, one report noted that no pretreatment factors were associated with an increased risk for the development of pulmonary toxicity.³⁵ Other studies have implicated factors such as pre-transplant restrictive lung disease, as measured by TLC <80% of predicted, and a DLCO <75% of predicted,³⁶ previous chest radiation,³⁴ and abnormal pre-treatment pulmonary function tests.³⁷ Each of these factors has been reported elsewhere to have no significant predictive value for the development of pulmonary toxicity during bone marrow transplantation.^38,39,40

As a result of the lack of clearly defined predictive variables, patients may unwisely receive high risk therapy, or alternatively may be excluded from a potentially life-saving procedure based on unproven judgement or dogma. We sought to better define these risk factors. We retrospectively reviewed the records of 138 patients who underwent autologous bone marrow transplantation from November 1990 to May 1995. Twenty-four relevant variables were chosen and analyzed for their predictive value in determining subsequent toxicity to the heart or lungs.

Patients and methods

Patients

Beginning in November 1990, all patients treated at our institution (and our affiliate Louis A Weiss Memorial Hospital) with HDC and ASCR for breast cancer and lymphoma were prospectively entered into a comprehensive data base. Patients treated on co-operative group studies and pharmaceutical company sponsored studies were not included in this database. From October 1990 until May 1995, data on 140 patients on three consecutive advanced breast cancer protocols and two lymphoma protocols were entered into the data base. All protocols were approved by the respective institutional review boards and informed consent was obtained from patients prior to treatment. Toxicity data were unavailable in two patients. Data were therefore available on 138 patients. The five high-dose chemotherapy regimens are described in Table 1. These regimens were used sequentially for lymphoma (protocols 6121, 5846) and metastatic breast cancer (protocols 5736, 7198, 6866). Notably, the regimens used were heterogeneous, although all breast cancer regimens used the same cyclophosphamide and thiotepa dose, and paclitaxel was added in one of three. One lymphoma regimen included busulfan, compared to none of the three breast cancer regimens.

Routine pretransplant organ evaluation

All patients had routine pretransplant testing performed. This included complete blood count (CBC) with differential and platelet count, chemistries, to include BUN and creatinine, and liver function tests (bilirubin, alkaline phosphatase, SGOT, SGPT, GGT) and serologies for HIV, hepatitis B and C (when available), HSV and CMV. Pulmonary function was evaluated with pulmonary function testing including spirometry, lung volumes, and diffusing capacity. Cardiac function was routinely evaluated with a first pass RNA or MUGA test, as below. In all cases additional data were obtained from history, physical examination, and X-rays done for other purposes. Patients with inadequate organ function, as defined below, were ineligible for transplant. Minor variations in these criteria existed between protocols: Serum creatinine >2.0 mg/dl, creatinine clearance <50 ml/min, FEV <1 L or FVC or FEV <65% predicted, DLCO <50% predicted, SGOT, SGPT, or alkaline phosphatase >2 ´ normal, or bilirubin >3.0 mg/dl.

Cardiology evaluation

Patients felt to be at high risk for cardiac toxicity from HDC were evaluated by one cardiologist (JAS), who also provided in-hospital follow-up. Indications for referral included low ejection fraction, abnormal EKG, or history of cardiac disease. Cardiology evaluation included a history and physical, EKG, first pass RNA or MUGA scan at rest and with exercise (when myocardial dysfunction was the issue for cardiology evaluation), and when necessary, an echocardiogram. A total of 19 patients were referred for cardiac evaluation, of whom 11 underwent HDC. The remaining eight patients will be described.

Prognostic variables

A total of 24 variables were rationally chosen for cardiac toxicity analysis (21 pre-transplant and three transplant-related variables). Sixteen variables were analyzed for pulmonary toxicity (Table 2). Data were obtained from our transplant patient database and augmented by chart review for some variables not included in the database. For each variable, data were available on most patients (Table 2). Data for smoking history and cardiac risk score ('Score' described below) were available in 124 and 110 patients, respectively.

The criteria used to categorize each variables are as follows: EKG readings were coded as normal, abnormal, or markedly abnormal. Abnormal finding (23 patients) included left ventricular hypertrophy (LVH), abnormal intervals, two or more PVCs or PACs on a standard 12 lead EKG, left axis deviation -30° (LAD), non-specific T-wave abnormalities, low voltage, or poor R-wave progression (PRWP). Criteria for markedly abnormal EKGs (four patients) were bundle branch block, evidence of myocardial infarction, or other arrhythmia. Because of the small number of patients meeting criteria for 'markedly abnormal', markedly abnormal patients were combined with abnormal patients.

Definition of previous cardiac disease included pericardial effusion, thickening, or other pericardial disease, valvular disease, coronary artery disease (CAD), arrhythmia, or myocardial disease. Excluded were minor abnormalities such as mitral valve prolapse, sinus tachycardia, and functional murmurs. Congestive heart failure (CHF) based on a low EF without clinical signs or symptoms was not included in this variable but was included as a separate category (see LEF, PEF, any EF below). No patients with clinical congestive heart failure underwent HDC or referral for cardiology consultation. Pre-existing cardiac disease existed only in five patients, all with pericardial disease.

We defined previous pulmonary disease to include prior pulmonary embolus (PE) or infarction, obstructive ventilatory defects, restrictive lung disease, significant parenchymal scarring, previous pneumothorax, prior drug toxicities, and other parenchymal and airways disease. Excluded were pleural effusions, pneumonia and metastatic lung nodules.

Cardiac risk score was calculated based on the Framingham Heart Study - coronary heart disease (CHD) risk prediction worksheet (see Appendix 1). The Framingham model takes into account age, sex, HDL cholesterol, total cholesterol, systolic blood pressure, smoking history, diabetes and EKG evidence of LVH. The score derived from an adaptation of this model was used as the 'score' variable. Because cholesterol levels in previously treated cancer patients may be inaccurate for a variety of reasons,^41,42,43 we used family history of CAD as a surrogate for cholesterol levels. This was defined as having one or both parents with, or having died from, CAD and was given a score of two points. Also, patients with less than a 10 pack year smoking history received no points. Those with any one of the variables missing did not receive a score. As a result, score was assigned to 110 of 138 patients.

Chest radiation was initially subdivided into RT to the right chest, left chest, bilateral chest, or mediastinum (excluding RT to the spine). Because there were too few occurrences of toxicity in any one category, these were subsequently combined into one variable (any history of chest radiation) to provide greater power to this variable.

Mobilization with cyclophosphamide variable distinguished patients who received cyclophosphamide (4 g/m²) in addition to G-CSF for stem cell mobilization.

Ejection fraction information included the lowest recorded EF (LEF) at any time prior to HDC, and the EF measured prior to treatment (PEF). EEF was the exercise ejection fraction measured in nine treated patients and described later (EEF obtained routinely in patients referred for cardiology evaluation due to suspected or measured myocardial dysfunction). An additional category, any EF, was coded for patients with any of LEF, EEF or PEF less than 50%. All measures of ejection fraction were performed with first pass RNA or MUGA scan.

Pack years of smoking was calculated from the average number of packs of cigarettes smoked per day multiplied by years of smoking (for example, an individual who smoked an average of one and a half packs of cigarettes daily for 20 years would have a 30 pack year history of smoking). A separate dichotomous variable, smoker accounted for patients with any history of cigarette smoking.

Total doses of doxorubicin (total doxorubicin), and cyclophosphamide (total cyclophosphamide) (in mg/m²) were taken from patient records. When exact drug dosages were not known, standard values for each drug were used for the specific regimens that the patient was known to have received. The dose of cyclophosphamide used for mobilization was included in total cyclophosphamide dose. Drug doses received during transplant were not included, as all patients received high-dose cyclophosphamide, all breast cancer patients additionally received high-dose thiotepa. Prior bleomycin usage was recorded as yes or no.

Disease (lymphoma vs breast), history of hypertension, history of lung metastases, pre-treatment performance status (PS), gender, age at transplant, DLCO (carbon monoxide diffusing capacity), FEV₁ (forced expiratory volume in 1 s), volume of reinfusion and number of chemo regimens (prior to mobilization) were taken from the data base or patient charts.

Toxicity criteria

The toxicity grading criteria used to evaluate toxicity were standard at the University of Chicago during the early 1990s, and had been proposed to be universally accepted ABMT criteria. These are shown in Table 3. For the purpose of this analysis, all toxicities were considered together regardless of grade.

Statistical methods

Logistic regression models⁴⁴ were used to model separately the probabilities of experiencing both cardiac and pulmonary toxicity as a function of various prognostic covariates. For each toxicity type, a model was fit including those covariates judged to be the most relevant clinically (those uniformly reviewed by oncologists prior to accepting a patient for HDC) and any other with P < 0.15 (P = 0.15 chosen to include 'statistically significant' variables and those which might become significant when controlling for other variables). These models were fit both with and without the covariates pack years and score, since data on these variables were missing in 14 and 28 cases, respectively. In addition, for each of the two types of toxicity, models were fit for all possible subsets of covariates. These models were then compared using the Akaike Information Criterion (AIC) in order to determine the 'best' prognostic model based on the data.⁴⁵ All effects are reported in terms of odds ratios together with approximate 95% confidence intervals (based on the standard normal approximation to the distribution of the parameter estimate, then transformed to the odds scale by exponentiating). For each of the continuous variables, smoothed plots of the partial residuals were examined to check for possible non-linearities in the effects of these variables.

Results

Distribution of variables

Univariate distributions of each of the covariates are described in Table 2. Slightly more than two-thirds of patients had breast cancer, and all of these were women. Cyclophosphamide was given as a component of mobilization chemotherapy in 53%. Smokers accounted for 41% of patients. All other dichotomous risk factors were present in a minority, ie 25% or less. Data for all of the covariates were present in at least 96% of patients with the exception of pack years (and smoker) with data on 124/138 (90%) and risk factor score (score) with data on 110/138 (80%). These two variables with missing data are therefore treated separately below.

Toxicity

Table 4 shows the incidence of both cardiac and pulmonary toxicity by disease type. Cardiac toxicity was experienced by 17 patients (12%) while 33 patients (24%) experienced pulmonary toxicity. Both types of toxicity occurred more frequently in lymphoma patients. Table 4 also demonstrates that these toxicities are roughly ordered; that is 14/17 patients (82%) experiencing cardiac toxicity also experienced pulmonary toxicity. Table 3 shows the breakdown of toxicity by ABMT criteria. Six patients with fatal (grade 4) cardiac toxicity also had fatal pulmonary toxicity. In four cases, congestive heart failure with objective moderate or severe decrease in cardiac output (on echocardiogram or right heart catheterization) was felt to be attributable to chemotherapy, specifically high-dose cyclophosphamide. Pulmonary toxicity in these four patients included hypoxia with suspected pulmonary hypertension, ARDS in the absence of sepsis in two, and capillary leak felt attributable to cytosine arabinoside in the other. A fifth patient had cardiac arrest with electromechanical dissociation just after stem cell reinfusion and had diffuse pulmonary infiltrates without pneumonia. The sixth patient had hypotension and capillary leak of unclear etiology and later developed pulmonary aspergillosis. In five cases, suspected or proven sepsis followed initial manifestations of cardiac and pulmonary toxicity and dysfunction, and in one case sepsis neither occurred nor was suspected. An additional two patients suffered fatal pulmonary toxicity. These deaths were due to ARDS without sepsis in one, and septic pulmonary emboli along with pulmonary Hodgkin's disease in the other.

Bivariate analysis

Table 2 shows the bivariate relationship between each of the covariates and the probability (odds ratio) of toxicity. Significant variables for the development of toxicity (95% confidence interval not crossing unity) included disease type (lymphoma) and score (high) for cardiac and cyclophosphamide for mobilization (absence of) for pulmonary toxicity.

Logistic regression models

Cardiac toxicity: Table 5 shows the results of multivariate models predicting the odds of cardiac toxicity. Logistic regression models were generated using bivariates significant at P 0.15. Model I includes all variables indentified in the bivariate analyses. EKG and PEF were added, although not significant separately, since abnormalities in these variables might trigger a cardiology consult or disqualification from a transplant protocol. In model I (Table 5), score was omitted due to missing data in 28/138 patients, but added back in model II. Thus 128/138 patients and 16/17 episodes of toxicity were included in model I, but only 105/138 patients and 13/17 episodes of toxicity were included in model II.

Model I demonstrates that only disease type (lymphoma) is a significant independent predictor of cardiac toxicity though EF 50 may increase risk. Adding score in model II (which eliminated data on 28 patients) changed model I only slightly. The positive association between score and toxicity seen in Table 2 vanished when controlling for other variables. When all subset models (excluding score) were considered (data not shown), the model including disease (lymphoma), age (older) and chest radiation (none) had the smallest value of AIC. Since the P value for chest irradiation was large (P = 0.17) and the estimated effect was in the 'wrong' direction, our best model includes only disease type and age. Table 6 shows the predicted probabilities of cardiac toxicity generated by this reduced model.

Pulmonary toxicity: The bottom panel of Table 5 shows models predicting the likelihood of pulmonary toxicity both with and without the variable pack years. Model I includes 127/138 patients and 31/33 episodes of toxicity and model II includes 113/138 patients and 25/33 episodes of toxicity. Logistic regression models were generated here using covariates significant at P < 0.15 level (cyclophosphamide for mobilization, pack years) and significant variables from model I for cardiac toxicity (disease type, age), described above. Previous pulmonary disease, history of bleomycin, FEVI and DLCO were added since these were felt to be strong potential risk factors for pulmonary toxicity and abnormalities in these variables might exclude patients from HDC.

Model I, without pack years, shows disease (lymphoma) and cyclophosphamide for mobilization (lack of) as the only significant variables. In model II, adding pack years to model I (which eliminated 14 patients with missing data) changed the model only minimally. Disease type remained near significant (lymphoma), FEVI became nearly significant, and pack years and cyclophosphamide for mobilization became insignificant. Thus, the seemingly paradoxical effect of cyclophosphamide mobilization being protective for pulmonary toxicity was probably related to the serendipitous relationship between smoking history and not receiving cyclophosphamide (smoking history did not preclude mobilization with cyclophosphamide).

Models were also fitted using all possible subsets of the covariates in model I. The model with disease type, age, and cyclophosphamide for mobilization had the lowest value of AIC. The effect of age was only minimally significant and cyclophosphamide for mobilization was both minimally significant and mechanistically unlikely. Therefore disease type remained the primary independent risk factor, and we did not generate predicted probabilities of toxicity from this model. Since the unique pulmonary toxicity in lymphoma patients is likely due to busulfan, received by 36 of 43 (84%) lymphoma patients, we concluded that no factor, other than the known pulmonary toxin busulfan, predicted accurately the occurrence of pulmonary toxicity.

Role of the HDC regimen

Patients were treated on five different chemotherapy regimens (Table 1). The individual protocols were not considered as covariate predictors due to low patient numbers on three of the protocols. The three breast cancer protocols, all containing at least cyclophosphamide and thiotepa, were condensed and compared with the two lymphoma protocols. The toxicities by protocol number are shown in Table 1. The multivariate models were repeated (data not shown) substituting protocol number for disease type. The results were not substantially different than those presented here (that is, substitution of conditioning regimen for disease type did not change the results).

Relationship of PEF, EF <50 and cardiac toxicity

The models in Table 5 for cardiac toxicity suggest a clear trend towards increasing toxicity with EF <50 while at the same time estimating a positive relationship between increasing PEF and risk. This seems counterintuitive, and Table 7 displays these data. The greatest risk for toxicity was not in the group with both PEF <50 and EF <50 (11% risk), but in the group with PEF 50 and any EF <50 (33% risk). Review of the 15 patients in this latter group revealed none for whom the increase in EF could be attributed to specific medical management such as diuretics, afterload reduction, or inotropes. Possible explanations are discussed below.

Isolated cardiac toxicity

Fourteen of 17 patients had pulmonary complications along with cardiac. We analyzed the other three to see if there were any factors isolated to cardiac toxicity. None were found. Two of the three had breast cancer; ages were 29, 49, 59; LEF was 47, 43, 68; PEF was 47, 52, 68; total doxorubicin dose was 390, 185, 240 mg/m².

Significance of prior doxorubicin dose

The total doxorubicin dose showed a trend towards significance as a univariate but was not a predictor in the multivariate model when analyzed as a continuous variable. Of interest, two of eight patients who had previously received >450 mg/m² doxorubicin developed cardiac toxicity (both grade 4, 483 mg/m², 733 mg/m²).

Role of cardiology referral

A total of 19 eligible patients for these protocols were referred for a cardiologist's opinion prior to HDC. Six patients were recommended to not undergo HDC for the following reasons: (5) poor left ventricular response to exercise (low EEF) (25%, 27%, 30%, 31%, 33%); (1) anterior wall motion abnormality (AWMA) on exercise MUGA test. Of these six patients, two underwent HDC off protocol, after informed consent of the risk, due to lack of other options. One had a resting EF of 49%, but EEF of 31%. The other had a resting EF of 38%, EEF of 40%, but an AWMA on exercise. Cardiac and pulmonary toxicity were experienced in zero and one patient, respectively. A total of 13 patients were 'cleared' for HDC, of whom nine underwent HDC and four did not, for a variety of reasons. In the five patients 'cleared' for HDC who were treated on protocol, cardiac and pulmonary toxicity occurred in one and two patients, respectively. The other four patients were treated off protocol using the same drug regimens and none experienced cardiac or pulmonary toxicity. These six total patients treated off protocol are not, therefore, represented in the regression analysis. Toxicities and correlation to LEF, EEF, and PEF can be seen in Table 8.

Discussion

We analyzed 138 consecutive cases of patients undergoing HDC with ASCR for lymphoma or breast cancer to determine factors that may predict risk for cardiac or pulmonary toxicity. We found that older age and lymphoma (vs breast cancer) were independent risk factors for the development of cardiac toxicity. We found no predictors of pulmonary toxicity other than lymphoma (vs breast cancer), likely due to exposure to busulfan, a known pulmonary toxin.

As with any study examining multiple potential predictive factors, our study may have been both underpowered to detect important true associations and may have happened upon 'significant' associations by chance. The latter was well exemplified by the apparently protective effect of the addition of cyclophosphamide to G-CSF for mobilization. The lack of power is reflected in the width of many of the confidence intervals. In addition, the applicability of these data may be limited since it may be relevant only to the pre-operative regimens used in this cohort of patients.

One of our primary objectives was to understand the role of the ejection fraction in determining risk for cardiac toxicity. We therefore examined three different ejection fraction parameters (LEF, PEF, any EF) and reviewed the correlation of exercise ejection fraction (EEF) with toxicity. We found that LEF had no correlation as a univariate, any EF appeared to predict cardiac toxicity, and that paradoxically, PEF trended in the 'wrong' direction. This dichotomy can be interpreted in several ways. One explanation is that ejection fraction has no effect on cardiac toxicity, at least amongst patients referred for HDC and not rejected prior to cardiology consultation. Selection bias may be an alternative explanation. Table 7 shows that patients with the greatest risk of cardiac toxicity have a history (any EF) of low EF but a high EF prior to HDC (PEF). These patients may have suboptimal myocardial function but escape the scrutiny given those patients with PEF <50. An additional explanation, not supported by our data, is that specific intervention with medication such as ACE inhibitors may increase the PEF but still leave patients vulnerable to cardiac toxicity. Finally, most of the patients with PEF >50% who experienced cardiac toxicity had an EF very near 50% which may, in those individuals, have represented a change from a higher baseline ejection fraction and therefore be indicative of previous myocardial damage. Regardless of the explanation, one can conclude that selected patients with a history of low ejection fraction, while at somewhat higher risk, can be safely treated with HDC if scrutinized by the medical oncologist and referred for a comprehensive assessment of cardiac health by a cardiologist. Based on the finding noted in Table 7, patients with a normal EF prior to HDC but a history of low EF appear to be at increased risk for cardiac toxicity and should be referred for cardiac evaluation. Moreover, those with PEF <50 may well have a good outcome and should not be uniformly excluded from HDC.

Lymphoma emerged from these data as the disease type most likely to be associated with both pulmonary and cardiac toxicity. This occurred despite controlling for differences in age, ejection fraction, chest irradiation, pulmonary function, performance status, and smoking history. The relationship to pulmonary toxicity is most likely due to busulfan, administered to 36 of 43 lymphoma patients. No clear explanation exists for cardiac toxicity, although lymphoma patients are probably more susceptible to infection, and may be more susceptible to immune-mediated enhancement of organ toxicity. Thus, an insult which might not manifest as toxicity in a breast cancer patient may manifest itself in a lymphoma patient. Although busulfan is not commonly cardiotoxic, it is possible that busulfan-related pulmonary toxicity predisposed patients to manifest cardiac toxicity (patients with pulmonary toxicity were more likely to experience cardiac toxicity).

Total dose of doxorubicin was not a contributory factor as a continuous variable in our model. It is noteworthy, however, that only 6/138 patients (4%) experienced grade 4 cardiac toxicity and that two of these occurred in the eight patients who had previously received >450 mg/m² of doxorubicin.

Our findings differ somewhat from that found by other investigators. Hertenstein et al²⁷ found no correlation between the development of major cardiac toxicity and the findings of a pre-transplant cardiac evaluation, including history, physical examination, rest and exercise EKG, chest X-ray, two-dimensional echocardiography and radionucleotide ventriculography. Minor cardiotoxic events were more common in patients with a reduced EF. Bearman et al²⁸ found a trend towards the development of cardiac toxicity in patients with a pretreatment ejection fraction less than 50%. In our study a history of low EF correlated to cardiac toxicity, although the pretransplant EF, taken by itself, did not. In contradistinction to other studies, we found no factor other than busulfan, or alternatively the diagnosis of lymphoma vs breast cancer, to predict for pulmonary toxicity.^34,35,36,37

Several conclusions should be drawn from our study. Older patients with lymphoma may be more susceptible to cardiac toxicity than younger patients and those with breast cancer. Patients with any history of low ejection fraction are probably at increased risk for cardiac toxicity. Patients who have received >450 mg/m² of doxorubicin or high doses of other anthracyclines may be susceptible to cardiac toxicity regardless of ejection fraction. In particular, patients with a normal EF prior to HDC but with a history of low ejection fraction may merit additional cardiac evaluation. Finally, a comprehensive cardiac evaluation including history and physical, EKG, exercise MUGA, and other tests as necessary, may help define candidates with apparently unfavorable risk factors who may have low risk of toxicity with HDC.

Acknowledgements

The authors gratefully thank LP Schumm and R Mick for their statistical analyses and advice, Sheila Dertz for her data management assistance, and Elis Perez for her secretarial assistance in the preparation of this manuscript.

1 Brockstein BE, Williams SF. High-dose chemotherapy with autologous stem cell rescue for breast cancer: yesterday, today and tomorrow. Stem Cells 1996; 14: 79-89, MEDLINE

2 Dillman RO, Barth NM, Nayak SK et al. High-dose chemotherapy with autologous stem cell rescue in breast cancer. Breast Cancer Res Treat 1996; 37: 277-289, MEDLINE

3 Vandenberghe EA, Goldstone AH. Autologous stem cell transplants in lymphomas. Ann Med 1996; 28: 137-144, MEDLINE

4 Stasi R, Venditti A, Del Poeta G et al. High-dose chemotherapy in adult acute myeloid leukemia: rationale and results. Leukemia Res 1996; 20: 535-549,

5 Droz JP, Culine S, Biron P et al. High-dose chemotherapy in germ-cell tumors. Ann Oncol 1996; 7: 997-1003, MEDLINE

6 Santos SW, Sensenbrenner LL, Bruke PJ et al. The use of cyclophosphamide for clinical marrow transplantation. Transplant Proc 1972; 4: 559-564, MEDLINE

7 Goldberg MA, Antin JH, Guinan EC et al. Cyclophosphamide cardiotoxicity: an analysis of dosing as a risk factor. Blood 1986; 68: 1114-1118, MEDLINE

8 Braverman AC, Antin JH, Plappert MT et al. Cyclophosphamide cardiotoxicity in bone marrow transplantation: a prospective evaluation of new dosing regimens. J Clin Oncol 1991; 9: 1215-1223, MEDLINE

9 Gottdiener JS, Appelbaum FR, Ferrans VJ et al. Cardiotoxicity associated with high-dose cyclophosphamide therapy. Arch Intern Med 1981; 141: 758-763, MEDLINE

10 Shan K, Lincoff AM, Young JB. Anthracycline-induced cardiotoxicity. Ann Intern Med 1996; 125: 47-58, MEDLINE

11 Bu'Lock FA, Mott MG, Oakhill A et al. Early identification of anthracycline cardiomyopathy: possibilities and implications. Arch Dis Child 1996; 75: 416-422, MEDLINE

12 Bristow MR, Thompson PD, Martin RP et al. Early anthracycline cardiotoxicity. Am J Med 1978; 65: 823-832, MEDLINE

13 Porembka DT, Lowder JN, Orlowski JP et al. Etiology and management of doxorubicin cardiotoxicity. Crit Care Med 1989; 17: 569-572, MEDLINE

14 Quezado ZM, Wilson WH, Cunnion RE et al. High-dose ifosfamide is associated with severe reversible cardiac dysfunction. Ann Intern Med 1993; 118: 31-36, MEDLINE

15 Kandylis K, Vassilomanolakis M, Tsoussis S et al. Ifosfamide cardiotoxicity in humans. Cancer Chemother Pharmacol 1989; 24: 395-396, MEDLINE

16 Collins C, Weiden PL. Cardiotoxicity of 5-fluorouracil. Cancer Treat Rep 1987; 71: 733-736, MEDLINE

17 Martin M, Diaz-Rubio E, Furio V et al. Lethal cardiac toxicity after cisplatin and 5-fluorouracil chemotherapy. Am J Clin Oncol 1989; 12: 229-234, MEDLINE

18 Rowinsky EK, Eisenhauer EA, Chaudhry V et al. Clinical toxicities encountered with paclitaxel (taxol). Semin Oncol 1993; 20: (Suppl. 3) 1-15, MEDLINE

19 Wood AJ. Paclitaxel (taxol). New Engl J Med 1995; 332: 1004-1014, MEDLINE

20 Mandel EM, Lewinski U, Djaldetti M. Vincristine-induced myocardial infarction. Cancer 1975; 36: 1979-1982, MEDLINE

21 Sivanesaratnam V. Mitomycin-C cardiotoxicity. Med J Aust 1989; 151: 300, MEDLINE

22 Schultz-Hector S. Radiation-induced heart disease: review of experimental data on dose response and pathogenesis. Int J Radiat Biol 1992; 61: 149-160, MEDLINE

23 Benoff LJ, Schweitzer P. Radiation therapy-induced cardiac injury. Am Heart J 1995; 129: 1193-1196, MEDLINE

24 Stewart JW, Fajardo LF, Gillette SM et al. Radiation injury to the heart. Int J Radiat Oncol Biol Phys 1995; 31: 1205-1211, MEDLINE

25 Lopez-Jimenez J, Cervero C, Munoz A et al. Cardiovascular toxicities related to the infusion of cryopreserved grafts: results of a controlled study. Bone Marrow Transpl 1994; 13: 789-793,

26 Yellowlees P, Greenfield C, McIntyre N. Dimethylsulphoxide-induced toxicity. Lancet 1980; 8202: 1004-1006,

27 Hertenstein B, Stefanic M, Schmeiser T et al. Cardiac toxicity of bone marrow transplantation: predictive value of cardiologic evaluation before transplant. J Clin Oncol 1994; 12: 998-1004, MEDLINE

28 Bearman SI, Petersen FB, Schor RA et al. Radionuclide ejection fractions in the evaluation of patients being considered for bone marrow transplantation: risk for cardiac toxicity. Bone Marrow Transplant 1990; 5: 173-177, MEDLINE

29 Durant JR, Norgard MJ, Murad TM et al. Pulmonary toxicity associated with bischloroethylnitrosourea (BCNU). Ann Intern Med 1979; 90: 191-194, MEDLINE

30 Blum RH, Carter SK, Agre K. A clinical review of bleomycin - a new antineoplastic agent. Cancer 1973; 31: 903-914, MEDLINE

31 Krowka MJ, Rosenow EC, Hoagland HC. Pulmonary complications of bone marrow transplantation. Chest 1985; 87: 237-246, MEDLINE

32 Chan CK, Hyland RH, Hutcheon MA. Pulmonary complications following bone marrow transplantation. Clin Chest Med 1990; 11: 323-332, MEDLINE

33 Twohig K, Matthay R. Pulmonary effects of cytotoxic drugs other than bleomycin. Clin Chest Med 1990; 11: 31-54, MEDLINE

34 Todd NW, Peters WP, Ost AH et al. Pulmonary drug toxicity in patients with primary breast cancer treated with high-dose combination chemotherapy and autologous bone marrow transplantation. Am Rev Resp Dis 1993; 147: 1264-1270, MEDLINE

35 Przepiorka D, Dimopoulos M, Smith T et al. Thiotepa, busulfan, and cyclophosphamide as a preparative regimen for marrow transplantation: risk factors for early regimen-related toxicity. Ann Hematol 1994; 68: 183-188, MEDLINE

36 Carlson K, Backlund L, Smedmyr B et al. Pulmonary function and complications subsequent to autologous bone marrow transplantation. Bone Marrow Transplant 1994; 14: 805-811, MEDLINE

37 Quakeck K. The lung as a critical organ in marrow transplantation. Bone Marrow Transplant 1994; 14: (Suppl. 4) S19-S28, MEDLINE

38 Hartsell WF, Czyzewski EA, Ghalie R et al. Pulmonary complications of bone marrow transplantation: a comparison of total body irradiation and cyclophosphamide to busulfan and cyclophosphamide. Int J Radiat Oncol Biol Phys 1995; 32: 69-73, MEDLINE

39 Nichols DG, Walker LK, Wingard JR et al. Predictors of acute respiratory failure after bone marrow transplantation in children. Crit Care Med 1994; 22: 1485-1491, MEDLINE

40 Jochelson M, Tarbell NJ, Freedman AS et al. Acute and chronic pulmonary complications following autologous bone marrow transplantation in non-Hodgkin's lymphoma. Bone Marrow Transplant 1990; 6: 329-331, MEDLINE

41 Behar S, Graff E, Reicher-Reiss H et al. Low total cholesterol is associated with high total mortality in patients with coronary heart disease. Eur Heart J 1997; 18: 52-59, MEDLINE

42 Bokemeyer C, Berger CC, Kuczyk MA, Schmoll HJ. Evaluation of long-term toxicity after chemotherapy for testicular cancer. J Clin Oncol 1996; 14: 2923-2932, MEDLINE

43 Fiorenza AM, Branchi A, Cardena A et al. Serum cholesterol levels in patients with cancer. Relationship with nutritional status. Int J Clin Lab Res 1996; 26: 37-42, MEDLINE

44 Hosmer DW Jr, Lemeshow, S. Applied Logistic Regression. John Wiley: New York, 1989,

45 Lawless JF, Singhal K. ISMOD: an all-subsets regression program for generalized linear models. I. Statistical and computational background. Comp Meth Prog Biomed 1987; 24: 117-124,

Table 9

Table 1  Conditioning regimens (HDC) and incidence of toxicity by regimen

Table 2  Sample distribution of predictor variables

Table 3  Toxicity grading scale and distribution of toxicities by ABMT grading criteria

Table 4 Distribution of cardiac and pulmonary toxicity by disease type

Table 5  Multivariate logistic regression models predicting occurrence of cardiac and pulmonary toxicity (estimated odds ratios and corresponding 95% confidence intervals)

Table 6  Predicted probabilities of cardiac toxicity, with 95% confidence intervals

Table 7  Relationship between any EF and PEF and cardiac toxicity

Table 8  Patients referred for cardiology consult who underwent HDC

Table 9 Coronary heart disease risk factor prediction worksheets (Score worksheet) (modified from Framingham heart study)