This analysis includes data from two population-based studies that were harmonized and pooled to create the Neighborhoods and Breast Cancer Study (NABC). The two studies include the San Francisco Bay Area Breast Cancer Study (SFBCS), a case–control study of breast cancer in African American, Hispanic, and non-Hispanic white women [18, 19]; and the Northern California site of the Breast Cancer Family Registry (NC-BCFR) [20, 21]. Both studies identified women newly diagnosed with a first primary invasive breast cancer through the Greater Bay Area Cancer Registry and screened cases by telephone to establish study eligibility and self-identified race/ethnicity (participation was 84 % in the SFBCS and 83 % in the NC-BCFR among cases contacted).

In SFBCS, eligible cases were aged 35–79 years who lived in Alameda, Contra Costa, San Mateo, San Francisco, or Santa Clara counties at the time of diagnosis. They included all Hispanics diagnosed between 1 April 1995 and 30 April 2002, all African Americans diagnosed between 1 April 1995 and 30 April 1999, and a random 10 % sample of non-Hispanic whites diagnosed between 1 April 1995 and 30 April 1999. Of 2,571 cases selected into the case–control study, 2,258 (88 %) completed the in-person interview.

In NC-BCFR, eligible cases were aged 18–64 years who lived in Alameda, Contra Costa, Marin, San Mateo, San Francisco, Santa Clara, Santa Cruz, and Monterey counties at the time of diagnosis. They included cases of any race/ethnicity diagnosed from 1 January 1995 to 30 September 1998; Hispanic, African American, and Asian American (Chinese, Filipina, and Japanese) cases diagnosed from 1 October 1998 to 30 April 2002; Hispanic and African American cases diagnosed from 1 May 2002 to 31 December 2008. Cases were enrolled in the NC-BCFR if they had indicators of increased genetic susceptibility [20, 21]. Cases not meeting these criteria were randomly sampled (2.5 % of non-Hispanic whites and 33 % of other racial/ethnic groups). Of 4,708 cases selected into NC-BCFR, 3,631 (77 %) completed the in-person interview.

For cases that participated in both studies (n = 339), we used data from the SFBCS interview. We limited our analytic sample to cases with a first primary invasive breast cancer, who completed the questionnaire themselves within 5 years of diagnosis, had a geocodeable address and had follow-up information from the cancer registry. We excluded cases with Native American or mixed race/ethnicity (n = 10) or unknown physical activity (n = 8). The remaining 4,345 cases were interviewed on average 20.2 months (standard deviation = 8.8 months; range = 0.3–60.0 months) after diagnosis. Study participants provided written informed consent and all protocols were approved by the Institutional Review Board of the Cancer Prevention Institute of California.

For both studies, trained interviewers administered similar, structured questionnaires at the participant’s home in English, Spanish, or Chinese. Data were harmonized according to common definitions and included age at diagnosis, race/ethnicity, education, first-degree family history of breast cancer, personal history of benign breast disease, years since last pregnancy, pre-diagnosis oral contraceptive use, pre-diagnosis menopausal hormone therapy use, grams per day of alcohol intake 1 year prior to diagnosis, and body mass index (BMI) 1 year prior to diagnosis.

Assessment of physical activity has been described elsewhere [8, 19]. Briefly, lifetime histories of strenuous and moderate recreational activities were available from both studies and combined for these analyses. SFBCS assessed lifetime recreational physical activity performed at least 1 h per week for at least 4 months per year; for each activity, information was collected on type of activity, the age the activity started and stopped, months per year performed, and hours per week performed. NC-BCFR assessed lifetime histories of moderate (e.g., brisk walking, hiking, cycling on level streets) and strenuous (e.g., swimming laps, aerobics, running, cycling on hills) activities; for each type of activity, information was collected on the average number of hours per week (0.5, 1, 1.5, 2, 3, 4–6, 7–10, ≥11, or do not know) and months per year (1–3, 4–6, 10–12, or don’t know) at specific ages, including during the 3 years before diagnosis. We estimated average recreational physical activity during the 3 years before diagnosis because prior research has shown associations between recent physical activity and survival [22], and this time frame corresponded most closely with the measurement of our neighborhood characteristics. Three physical activity variables were considered. For the first variable, the average hours per week and months per year of moderate and strenuous physical activities during the 3 years before diagnosis were summed and categorized into the following: (1) meeting physical activity recommendations by the American Cancer Society (ACS) [23] (at least 150 min of moderate intensity, 75 min of strenuous intensity, or an equivalent combination of moderate/strenuous activity per week); (2) not meeting the recommendations, but performing some moderate or strenuous physical activity; or (3) performing no moderate or strenuous physical activity. For the second variable, recent moderate and strenuous activities were weighted by metabolic equivalents (MET) (8.5 for strenuous activity and 5.4 for moderate activity) [24] and summed to obtain MET hours per week of activity during the 3 years before diagnosis, as done previously [8]. For the third variable, we considered hours per week of moderate and strenuous activity not weighted by MET.

For each participant, we obtained cancer registry information routinely abstracted from the medical record at diagnosis [25], including tumor histological subtype (ductal, lobular, or mixed/other), histological grade, estrogen receptor (ER) and progesterone receptor (PR) status, AJCC (American Joint Committee on Cancer) stage, time to first and second subsequent tumors, first-course treatment (chemotherapy, radiation, and surgery), marital status and vital status (routinely determined by the cancer registry through hospital follow-up and database linkages) as of 31 December 2009 and, for the deceased, the underlying cause of death.

Residential address at the time of diagnosis was geocoded to a latitude/longitude coordinate and then assigned a 2000 census block group. Addresses were standardized to conform to U.S. Postal Service specifications using ZP4 software (ZP4. Monterey, CA: Semaphore Corp., 2011). Batch geocoding was performed using ArcGIS with both current address point and street geocoding reference files (ArcGIS. Redlands, CA: Environmental Systems Research Institute, Inc., 2011). Manual review was performed to geocode addresses that did not batch geocode, resulting in 97 % of addresses being assigned a latitude and longitude.

For neighborhood-level socioeconomic status (SES), we used a previously validated composite SES measure of seven indicator variables at the census block group (education index, median household income, percent living 200 % below poverty level, percent blue-collar workers, percent older than 16 in workforce without job, median rent, median house value) [26]. Neighborhood immigrant population was characterized at the block group level with percentage of foreign-born residents.

We derived information on neighborhood amenities including business listings from Walls & Associates’ National Establishment Time-Series Database (which utilizes data from Dunn and Bradstreet) [27], farmers’ markets listings from the California Department of Food and Agriculture [28], and parks from the NavTeq’s NavStreets database [29]. Using ArcGIS software, neighborhood amenities within a 1,600-m network distance [30] from a case’s residence at diagnosis were summed. To do this, we first constrained the search space by selecting all neighborhood amenities within a 1,600-m linear distance of residence, then computed the actual network distance between the residence and every neighborhood amenity. If the network distance was ≤1,600 m, then the neighborhood amenity was included in analyses. We considered businesses active 1 year before diagnosis, during the year of diagnosis, or 2 years after diagnosis. The number of recreational facilities included places where recreational activities could take place (e.g., fitness centers). The Restaurant Environment Index is the ratio of the number of fast-food restaurants to other restaurants, and the Retail Food Environment Index [31] is the ratio of the number of convenience stores, liquor stores, and fast-food restaurants to supermarkets and farmers’ markets. Parks included beaches, recreation areas, and parks. We determined the number of parks with an access point (e.g., vehicular entry point/parking lot or edge of smaller park) within 1,600-m network distance of a participant’s residence.

Neighborhood density was characterized at the census block group level by population density (the number of people/m²) and percentage of total housing units that are not single-family dwellings. Population-based commuting characteristics from the census were summarized by typical travel time to work (% traveling ≥60 min/day). Data on traffic counts from the California Department of Transportation [32] were used to obtain traffic density within a 500-m buffer of each case’s residence, using methods described previously [33].

To evaluate the association of personal and neighborhood characteristics with physical activity guidelines [23], we used logistic regression to calculate odds ratios (ORs) and 95 % confidence intervals (CIs). To evaluate the association of physical activity and neighborhood characteristics with overall and breast cancer-specific survival, we used Cox proportional hazards regression to calculate hazard ratios (HRs) and 95 % CI. Base models for both logistic and Cox proportional hazards regression models included age at diagnosis, study, race/ethnicity, stage at diagnosis, and clustering by block group. Neighborhood factors that were associated with survival or physical activity in the base models were included in the fully adjusted model. Logistic models also included potential confounding variables significantly associated with physical activity (education, pre-diagnosis BMI, pre-diagnosis menopausal hormone therapy use, pre-diagnosis alcohol intake). In the fully adjusted Cox models, the confounding variables included personal factors (marital status, education, history of benign breast disease, years since last full-term pregnancy, pre-diagnosis oral contraceptive use, pre-diagnosis menopausal hormone therapy use, pre-diagnosis alcohol intake, pre-diagnosis BMI), tumor characteristics, and treatment. Correlation among the neighborhood variables was assessed, and only uncorrelated neighborhood factors (<0.50 Spearman correlation coefficient) that were significantly associated with the outcomes in univariate models were included in the multivariate models. There were insufficient numbers of cases within each neighborhood unit (block group) to warrant multilevel modeling (55 % of block groups had one case and 80 % of block groups had two or fewer cases).

For deceased cases, survival time was measured in days from the date of interview to the date of death of any cause for overall survival and to the date of death from breast cancer for breast cancer-specific survival. For breast cancer-specific survival, patients who died from other causes were censored at the time of death. Patients alive at the study end date (31 December 2009) were censored at this date or at date of last follow-up (i.e., last known contact). The proportional hazards assumption was tested for physical activity and neighborhood variables using significance tests of interactions with the timescale, and visual examination of scaled Schoenfeld residual plots; there was no evidence that these variables violated the assumption of proportional hazards.

We performed stratified analyses according to age at diagnosis (<50 or ≥50 years), BMI (<25.0, 25.0–29.9, ≥30 kg/m²), ER status (ER+, ER−, unknown), and race/ethnicity. Tests for heterogeneity across strata were conducted using likelihood ratio tests comparing models with and without an interaction term between physical activity and the stratified variable; no significant interactions were found (data not shown). Tests for trend were used to evaluate associations between survival and increasing physical activity and ordinal categories of neighborhood characteristics. p values<0.05 were considered statistically significant, and all tests of significance were two-sided. Analyses were performed using SAS version 9.3 (SAS Institute, Cary, NC).

In the fully adjusted model, not meeting recreational activity recommendations was associated with diagnosis with AJCC stage IV disease, Hispanic and Asian American race/ethnicity, lower education, being obese, and residence in neighborhoods with a high percentage of foreign-born residents, no fast-food restaurants (vs. fewer fast-food restaurants) to other restaurants, and high traffic density (Table 2). There was also borderline significant evidence that meeting recreational physical activity recommendations was associated with being overweight and residing in neighborhoods without convenience or liquor stores and fast-food restaurants (vs. neighborhoods with all types of retail food environment, including supermarkets and farmer’s markets). Number of parks and recreational facilities were not associated with physical activity levels in the fully adjusted model.

After adjustment for personal, tumor, treatment, and neighborhood characteristics (Table 3), Hispanic and Asian American women had better survival than non-His-panic white women, whereas African American and non-Hispanic white women had similar overall survival. Women who were not recreationally physically active in the 3 years prior to diagnosis had a 22 % HR 1.22, 95 % CI 0.98–1.52) higher risk of death from any cause than women that were the most active. Results were similar and statistically significant when considering hours per week of moderate or strenuous physical activity not weighted for MET [fully adjusted HR 1.27 (95 % CI 1.02–1.59); comparing women with no physical activity to those with the most activity] and meeting physical activity recommendations [fully adjusted HR 1.22 (95 % CI 1.03–1.44) comparing women with no physical activity to those meeting physical activity recommendations] (data not shown in tables). Poorer overall survival was associated with lower neighborhood SES (p trend = 0.02) and residence in more densely populated neighborhoods, although the trend for population density was not significant in the fully adjusted model. Associations of better overall survival with residence in neighborhoods with no parks were only marginally significant in the fully adjusted models. Additional analyses stratified by neighborhood SES showed that the inverse association was present only among women who resided in high-SES (quintiles 4, 5) neighborhoods [HR 0.68 (95 % CI 0.49–0.96) comparing no parks vs. ≥3 parks], while no association was seen among women in lower-SES (quintiles 1–3) neighborhoods [HR 1.05 (95 % CI 0.67–1.25)] (data not shown in tables).

Although the associations with neighborhood characteristics were generally similar for overall and breast cancer-specific survival, there was no association between recreational physical activity and breast cancer-specific survival (Table 4). Results were similar for hours per week of moderate or strenuous physical activity not weighted by MET [fully adjusted HR 1.04 (95 % CI 0.80–1.37) comparing women with no physical activity to those with the most activity] and meeting physical activity recommendations [fully adjusted HR 1.06 (95 % CI 0.85–1.31) comparing women with no physical activity to those meeting physical activity recommendations] (data not shown in tables). Hispanic women had better survival than non-Hispanic white women. Although the trends were not statistically significant, residence in more densely populated neighborhoods was associated with poorer survival, whereas residence in neighborhoods with more foreign born was associated with better survival. Living in neighborhoods without parks was associated with better breast cancer-specific survival. As with overall survival, the association was only seen among women living in high-SES neighborhoods [HR 0.54 (95 % CI 0.34–0.85) comparing no parks vs. ≥3 parks], while no association was seen among women living in lower-SES neighborhoods [HR 1.28 (95 % CI 0.52–1.21)] (data not shown in tables).