Risk Factors for Lymphedema in a Prospective Breast Cancer Survivorship Study:  The Pathways Study FREE

Breast cancer survival rates have been increasing during the past decade,¹ resulting in more long-term health consequences of breast cancer treatment. With incidence rates of approximately 26% 2 years after surgery, at least 400 000 women may be currently living with breast cancer–related lymphedema (BCRL),² the disruption of lymph transport leading to constant buildup of protein-rich fluid within the subcutaneous tissue of the arm, hand, and/or chest.^3- 5 Possible risk factors include the degree of axillary dissection, radiation therapy, obesity at diagnosis, older age, postoperative fluid formation, and infection in the arm.⁶

Reports of BCRL incidence vary widely owing to nonstandard diagnostic methods, type of cancer therapy, and follow-up after treatment.⁴ Incidence ranges from 10% to 56% after axillary lymph node dissection (ALND), 0% to 23% after sentinel lymph node biopsy (SLNB), and 21% to 51% after axillary radiation therapy and lymph node surgery.^2,7 The prevalence increases with time since treatment.^8- 9 In addition, BCRL can cause functional impairment of the affected arm^10- 11 and psychological morbidity,^12- 14 both of which can lead to an overall decreased quality of life.^9,15- 18

Given the limited number of prospective studies examining determinants of BCRL, we initiated a prospective cohort study among patients recently diagnosed as having breast cancer within Kaiser Permanente Northern California (KPNC). The Pathways Study is one of the first to systematically examine the incidence of BCRL as well as demographic, lifestyle, and clinical factors associated with its development. We present data from the first 997 enrolled participants with a mean follow-up time of 20.9 months.

The Pathways Study is an ongoing prospective cohort study actively recruiting women from the KPNC patient population recently diagnosed as having invasive breast cancer. As of January 30, 2009, 2415 patients have been enrolled since recruitment began on January 9, 2006, with a mean time from breast cancer diagnosis to enrollment of 1.9 months (range, 0.3-7.3 months). Participants are largely representative of the overall KPNC breast cancer population yet are slightly younger and more likely to be diagnosed as having earlier-stage disease, as described elsewhere.¹⁹ Cases are rapidly ascertained on a daily basis by automatic scanning of electronic pathologic reports, with subsequent verification of cancer diagnosis and patient notification by a medical records analyst. Eligibility criteria include current KPNC membership, age of at least 21 years at diagnosis, recent diagnosis of primary invasive breast cancer (all stages), no history of any cancer, and ability to speak English, Spanish, Cantonese, or Mandarin. Written informed consent is obtained from participants before commencement of the in-person interview. The study was approved by the institutional review boards of all collaborating institutions.

Diagnosis of BCRL requires consideration of multiple clinical factors, including postsurgical transient swelling (not considered lymphedema). The classic manifestation of lymphedema is defined by swelling.²⁰ However, more recent guidelines in the National Cancer Institute Common Terminology Criteria for Adverse Events version 3.0²¹ acknowledge that lymphedema includes not only swelling but also tissue tone and texture. Diagnosis requires examination by a specialist trained as recommended by the National Lymphedema Network.²²

Using KPNC electronic medical records, a diagnosis of lymphedema was determined from 3 sources: (1) outpatient or hospitalization International Classification of Diseases, Ninth Revision,²³ diagnostic codes 457.0 (postmastectomy lymphedema syndrome) and 457.1 (other lymphedema, including acquired [chronic]); (2) outpatient procedural codes corresponding to lymphedema treatment and duration of the procedure, such as manual lymph drainage (Current Procedural Terminology -4 code 97140 and internal KPNC qualifiers signifying duration) and compression wrap (Current Procedural Terminology -4 code 97139 [https://catalog.ama-assn.org/Catalog/cpt/cpt_search.jsp] and internal KPNC qualifiers signifying duration); and (3) durable medical equipment (DME) orders associated with BCRL. Relevant events must have occurred after the primary breast cancer diagnosis. Women were excluded if they had a history of lymphedema diagnosis, procedures, or DME orders.

For the analysis, 3 definitions of a BCRL diagnosis were developed: (1) transient and persistent lymphedema: an outpatient clinic or hospitalization diagnosis, BCRL procedure, or DME order anytime after breast cancer diagnosis, (2) persistent lymphedema: an outpatient clinic or hospitalization diagnosis, BCRL procedure, or DME order at least 3 months after definitive breast cancer surgery, and (3) persistent lymphedema: an outpatient clinic or hospitalization diagnosis anytime after definitive breast cancer surgery followed by a BCRL procedure or DME order.

During the interview, demographic information (age at breast cancer diagnosis, race/ethnicity, educational level, marital status, and annual household income) as well as lifestyle factors (body mass index [BMI; calculated as weight in kilograms divided by height in meters squared], smoking status, and physical activity) were obtained. The BMI was calculated from self-reported weight and height, and missing values were supplemented by concurrent data from KPNC electronic medical records. Physical activity in metabolic equivalent tasks (hours per week) was assessed from a previously validated, self-administered questionnaire.²⁴

Using KPNC electronic medical records, data on breast surgery (lumpectomy and mastectomy) and lymph node surgery (ALND and SLNB) were obtained using International Classification of Diseases, Ninth Revision²³ (85.20-85.23, 85.33-85.48, 40.23, 40.3, and 40.51) and Current Procedural Terminology -4 (19120-19240, 19301-19307, 19340-19342, 38500-38530, 38740, 38745, and 38792) diagnostic codes. Number of lymph nodes removed during the definitive surgery was abstracted from pathology reports by a trained staff member (R.C.). Data on number of positive nodes, American Joint Committee on Cancer (AJCC) stage, and chemotherapy and radiation therapy were obtained from the KPNC Cancer Registry.²⁵ Data were collected, coded, and added to the KPNC Cancer Registry approximately 4 months after diagnosis to allow for the completion of treatment. Radiation therapy data were supplemented by other electronic sources. Information on adjuvant hormone therapy was abstracted from outpatient pharmacy records.

Our analysis is based on the first 997 women enrolled in the Pathways Study with complete demographic, lifestyle, and clinical data. Relevant characteristics, as determined a priori from previous literature and current hypotheses, between women who developed BCRL and those who did not were compared using Pearson χ² tests (categorical variables) and Satterthwaite t tests (continuous variables). Follow-up began on the date of initial breast cancer diagnosis and continued until BCRL diagnosis or July 31, 2008, whichever occurred first; women were censored on the date of death or disenrollment from the KPNC medical care program (defined as first occurrence of a gap in coverage greater than 90 days).

The Kaplan-Meier method was used to generate a cumulative incidence curve²⁶ from initial breast cancer diagnosis to occurrence of BCRL in the cohort by plotting the complement of the survival function estimates. Initially, univariate Cox proportional hazards models²⁷ were constructed to calculate hazard ratios (HRs) and 95% confidence intervals (CIs) for each selected demographic, lifestyle, and clinical factor and risk of BCRL as defined previously. Subsequently, only variables that retained a priori interest and/or statistical significance were included in 3 final multivariate models, with each model estimating the associations between the selected covariates (adjusted for each other) and 1 of the 3 definitions of BCRL diagnosis. Linear tests of trend were estimated by modeling continuous predictor variables (age, BMI, and physical activity) and categorical predictor variables (household income) on an ordinal scale. All analyses were performed using SAS version 9.1 statistical software (SAS Institute, Cary, North Carolina).

Table 1 gives selected demographic, lifestyle, and clinical characteristics among 133 patients with BCRL and 864 patients without BCRL in the cohort. Women with BCRL were younger (55.3 years vs 60.1 years), more educated (58.6% vs 44.9% had completed college and/or graduate school), and were less likely to have smoked (63.9% vs 50.6%) compared with patients without BCRL. Patients with BCRL had more advanced cancer at diagnosis (AJCC stage II or III); were more likely to have had a mastectomy (58.6% vs 34.6%), ALND (86.5% vs 72.1%), or SLNB (84.2% vs 71.4%); and were more likely to have had radiation therapy (77.4% vs 70.3%) or chemotherapy (74.4% vs 39.7%) compared with patients without BCRL. On average, women with BCRL had had more lymph nodes removed (12.4 vs 6.1) and more positive lymph nodes (3.3 vs 0.8) compared with women without BCRL. No significant differences were observed between the 2 groups according to race/ethnicity, marital status, household income, BMI, physical activity, or adjuvant hormone therapy.

Among the 997 women who received follow-up for a mean period of 20.9 months (range, 0.7-31.8 months) after breast cancer diagnosis, 133 women (13.3%) had a clinical indication for BCRL with a mean time to diagnosis of 8.3 months (range, 0.7-27.3 months) (Figure). At 12 and 24 months after breast cancer diagnosis, the cumulative incidence of BCRL was 10.4% and 13.5%, respectively.

Table 2 presents the associations between demographic, lifestyle, and clinical characteristics and risk of BCRL, adjusted for all characteristics of interest in 3 multivariate proportional hazards models. For model 1, which determined the risk of transient and persistent BCRL, African American women had an increased risk of BCRL compared with white women (HR, 1.93; 95% CI, 1.00-3.72). The risk appeared to be increased in other minority groups as well but was not statistically significant (Asian Americans: HR, 1.65; 95% CI, 0.88-3.10; other races or ethnicities: HR, 1.77; 95% CI, 0.73-4.25). Compared with women who had a high school education or less, women with a college education or higher had an increased risk of BCRL (college education: HR, 2.58; 95% CI, 1.18-5.63; postgraduate education: HR, 2.72; 95% CI, 1.21-6.11; P for trend = .03). Higher BMI was suggestive of an increased risk of BCRL among obese women (HR, 1.43; 95% CI, 0.88-2.31), yet results of the test for trend were not significant. Overall physical activity and moderate to vigorous physical activity were not associated with risk of BCRL.

Among the clinical characteristics, women diagnosed as having a more advanced cancer stage had an increased risk of BCRL (AJCC stage II: HR, 2.58; 95% CI, 1.43-4.65; AJCC stage III: HR, 5.39; 95% CI, 2.47-11.75). In addition, for each lymph node removed, BCRL risk increased by 4.1% (corresponding to a 49.3% increase in risk for every 10 nodes removed). Elevated yet nonsignificant associations were observed for mastectomy (HR, 1.29; 95% CI, 0.77-2.16) and lymph node dissection (axillary: HR, 1.37; 95% CI, 0.44-4.28; sentinel: HR, 1.54; 95% CI, 0.49-4.86; axillary and sentinel: HR, 1.75; 95% CI, 0.62-4.94). Chemotherapy (HR, 1.38; 95% CI, 0.81-2.38) and radiation therapy (HR, 1.25; 95% CI, 0.70-2.25) were associated with modestly increased risks of BCRL, whereas hormone therapy was not associated with risk (HR, 0.96; 95% CI, 0.62-1.47).

Similar associations with generally wider CIs were observed in models 2 and 3 regarding risk of persistent BCRL compared with model 1, although some differences were apparent (Table 2). In model 2, chemotherapy was significantly associated with risk of BCRL (HR, 2.04; 95% CI, 1.03-4.03) compared with model 1 (HR, 1.38; 95% CI, 0.81-2.38). For model 3, being obese was significantly associated with an increased risk of BCRL (HR, 2.34; 95% CI, 1.23-4.44) compared with model 1 (HR, 1.43; 95% CI, 0.88-2.31).

In this prospective cohort study of 997 breast cancer survivors who received follow-up for BCRL development for a mean period of 20.9 months, 133 women (13.3%) developed BCRL—as defined by a clinical diagnosis, procedure, or DME order—with a mean time to diagnosis of 8.3 months. Being African American and more educated was associated with an increased risk of BCRL. Advanced-stage breast cancer and number of lymph nodes removed were also associated with an increased risk of BCRL, although no significant association was observed for mastectomy and type of lymph node surgery. Smoking status and physical activity were not associated with BCRL risk, but being obese was suggestive of an elevated risk. After restricting BCRL diagnoses to exclude any transient swelling that might be treated as BCRL, all of the associations remained, with a stronger association of BCRL with obesity and chemotherapy.

Our results confirm and contrast with findings from 2 recent studies of BCRL.^6,28 Paskett et al⁶ conducted a prospective study of 622 young breast cancer survivors (≤45 years at diagnosis) using semiannual questionnaires and assessed factors associated with incidence of arm and/or hand swelling using adjusted Cox proportional hazards models. Similar to our results, the authors found that lymph node removal was associated with an increased risk of swelling (for each node removed, HR, 1.02; 95% CI, 1.01-1.04). They observed that receiving chemotherapy (HR, 1.76; 95% CI, 1.10-2.82), being obese (HR, 1.51; 95% CI, 1.09-2.09), and being married (HR, 1.36; 95% CI, 1.00-1.85) were also associated with an elevated risk, whereas age at breast cancer diagnosis (HR, 1.00; 95% CI, 0.97-1.03) and radiation therapy (HR, 1.23; 95% CI, 0.86-1.77) were not associated with risk. With the exception of marital status, those findings are similar to ours.

Hayes et al²⁸ evaluated 287 women between 6 and 18 months after breast cancer treatment for BCRL using bioimpedance spectroscopy and assessed the associations of personal, treatment, and behavioral characteristics with BCRL risk in adjusted logistic regression models. Similar to our study, they observed that removal of lymph nodes was associated with an increased risk (odds ration [OR], 3.9; 95% CI, 0.5-28.9) and that radiation therapy was not associated with risk (OR, 0.9; 95% CI, 0.2-4.3). In contrast to our null results, Hayes and colleagues observed that lower income (<$52 000 per year) was associated with a decreased risk (OR, 0.2; 95% CI, 0.1-0.5), whereas having a mastectomy (OR, 5.9; 95% CI, 1.4-22.5) and being at least 50 years old at cancer diagnosis (OR, 3.3; 95% CI, 1.0-11.1) were associated with an increased risk. They also reported that leading a sedentary lifestyle (OR, 6.1; 95% CI, 1.3-27.6) was associated with an increased risk of BCRL, thus suggesting, along with several studies observing no increase in symptoms after participating in various exercise programs,^29- 33 that it is safe for women with BCRL to engage in physical activity.

The generally consistent findings across these studies and ours suggest that educational attainment, BMI, and lymph node removal may be risk factors for BCRL. The association of greater BCRL risk with higher educational level may be due to increased clinical consultation for swelling by more educated women, thereby resulting in additional BCRL diagnoses. Weight status at breast cancer diagnosis might be a predictor of developing BCRL, also suggested by other studies^6,34- 36; obesity can contribute to an increased risk of postoperative complications and reduced muscle-pumping efficiency within loose tissues.^34,37 As for chemotherapy, glucocorticoid agents such as dexamethasone can produce the adverse effect of swelling,^38- 39 which may also increase BCRL risk.

Removal of lymph nodes, rather than type of lymph node surgery, may be more germane to the risk of BCRL. Observational studies have documented a lower risk of BCRL and other arm morbidity symptoms among patients who underwent SLNB only compared with ALND.^40- 46 For example, Francis et al⁴⁶ observed a 16.8% incidence of BCRL after SLNB using the National Cancer Institute Common Terminology Criteria for Adverse Events version 3.0 criteria for diagnosis compared with an increased risk of 47.1% after ALND. In contrast, no difference in risk at 1 year of follow-up was observed between women who underwent SLNB and ALND compared with SLNB alone in the American College of Surgeons Oncology Group randomized clinical trial.⁴⁷ Final results from the National Surgical Adjuvant Breast and Bowel Project clinical trial of more than 5500 patients with breast cancer are still outstanding.⁴⁸ Although the evidence suggests that SLNB compared with ALND is associated with a lower risk of BCRL,⁴⁹ likely owing to fewer lymph nodes removed with SLNB, we found that even with 1 removed node, a woman is at an increased risk (4.1%) for BCRL. Similarly, in a retrospective study of 1338 older patients with breast cancer who were diagnosed as having in situ or invasive cancer and self-reported having BCRL, an increased risk of BCRL was observed with successive removal of nodes but not according to type of surgery.⁵⁰

To our knowledge, ours is the first study to report an elevated risk of BCRL among minority groups, specifically African Americans, independent of confounding factors such as age at diagnosis, breast cancer stage, and socioeconomic status (educational level and household income). Although we cannot completely rule out residual confounding and spurious results due to small sample size in our analysis, 1 retrospective study noted a higher prevalence of BCRL among women from minority backgrounds.⁵¹ This racial/ethnic difference is consistent with the hypothesis that, compared with white women, women from minority groups may be more likely to do manual labor involving the upper body and to return to work before full recovery and/or arm rehabilitation after their breast cancer surgery.

Strengths of this study include being one of the few large prospective cohort studies of BCRL and being conducted within KPNC, one of the largest health maintenance organizations in the United States with electronic medical records, facilitating rapid case ascertainment, participant follow-up, and identification of BCRL. We assessed BCRL by clinical indication using standard diagnostic and procedural codes. Furthermore, as part of a prospective study of breast cancer survivorship, we were able to examine the associations between demographic, lifestyle, and clinical characteristics and risk of BCRL.

Several limitations should be noted. First, we could examine only the effect of overall radiation therapy, rather than radiation therapy to the axilla, owing to a lack of information on the anatomical site of treatment in our electronic data sources. Second, considering our outcome ascertainment methods using electronic medical records, we could have missed underlying BCRL cases not brought to the attention of, or not diagnosed by, a health care professional and therefore not documented in the data sources. Conversely, we could have included BCRL cases diagnosed for preventive or evaluative purposes only, although when we limited our definition of BCRL diagnosis to outcomes of likely persistent BCRL, most of the associations remained.

In conclusion, our findings confirm the results of previous studies, especially those concerning the increased risk of BCRL with removal of lymph nodes and being obese. Although we did not examine change in BMI, management of body weight may be one avenue for decreasing the risk of BCRL. We also observed an increased risk among minority groups, specifically African American women, compared with white patients, and reasons for these differences need to be pursued. Considering the functional and psychosocial effects of developing BCRL, instituting educational programs that include a detailed clinical profile of identified risk factors before surgery might lead to improved prevention and treatment of this debilitating condition.

Correspondence: Marilyn L. Kwan, PhD, Division of Research, Kaiser Permanente Northern California, 2000 Broadway, First Floor, Oakland, CA 94612 (Marilyn.L.Kwan@kp.org).

Accepted for Publication: September 2, 2009.

Author Contributions:Study concept and design: Kwan and Kushi. Acquisition of data: Kwan, Darbinian, Citron, Partee, and Kushi. Analysis and interpretation of data: Kwan, Darbinian, Schmitz, Kutner, and Kushi. Drafting of the manuscript: Kwan and Darbinian. Critical revision of the manuscript for important intellectual content: Kwan, Darbinian, Schmitz, Citron, Partee, Kutner, and Kushi. Statistical analysis: Kwan and Darbinian. Obtained funding: Kwan and Kushi. Administrative, technical, and material support: Kwan, Darbinian, Citron, Partee, Kutner, and Kushi. Study supervision: Kwan and Kushi.

Financial Disclosure: None reported.

Funding/Support: This study was supported by grant R01 CA105274 from the National Cancer Institute (Dr Kushi) and grant RSG-06-209-01-LR from the American Cancer Society (Dr Kwan).

Disclaimer: The contents of this manuscript are solely the responsibility of the authors and do not necessarily represent the official views of the funding agencies.

Additional Contributions: We gratefully acknowledge Bruce Fireman, MA, for biostatistical support; Erin Weltzien, BA, and Isaac Joshua Ergas, MPH, for programming assistance; the Pathways Study office and field staff; and most important, the Pathways Study participants.