Risk factors for delay of adjuvant chemotherapy in non-metastatic breast cancer patients: A systematic review and meta-analysis involving 186982 patients

Purpose Delay performance of adjuvant chemotherapy (AC) after surgery has been presented to affect survival of breast cancer patients adversely, but the risk factors for delay in initiation remain controversial. Therefore, we conducted this systematic review of the literature and meta-analysis aiming at identifying the risk factors for delay of adjuvant chemotherapy (DAC) in non-metastatic breast cancer patients. Methods The search was performed on PubMed, Embase, Chinese National Knowledge Infrastructure and Wanfang Database from inception up to July 2016. DAC was defined as receiving AC beyond 8-week after surgery. Data were combined and analyzed using random-effects model or fixed-effects model for risk factors considered by at least 3 studies. Heterogeneity was analyzed with meta-regression analysis of year of publication and sample size. Publication bias was studied with Egger’s test. Results A total of 12 observational studies including 186982 non-metastatic breast cancer patients were eligible and 12 risk factors were analyzed. Combined results demonstrated that black race (vs white; OR, 1.18; 95% CI, 1.01–1.39), rural residents (vs urban; OR, 1.60; 95% CI, 1.27–2.03) and receiving mastectomy (vs breast conserving surgery; OR, 1.35; 95% CI, 1.00–1.83) were significantly associated with DAC, while married patients (vs single; OR, 0.58; 95% CI, 0.38–0.89) was less likely to have a delay in initiation. No significant impact from year of publication or sample size on the heterogeneity across studies was found, and no potential publication bias existed among the included studies. Conclusions Risk factors associated with DAC included black race, rural residents, receiving mastectomy and single status. Identifying of these risk factors could further help decisions making in clinical practice.


Introduction
Breast cancer is the most common type of malignant tumor and second leading cause of death in women worldwide. It is estimated that there will be 249,260 new cases and 40,890 deaths in United States in 2016 [1], which places a heavy burden on the healthcare system. Surgery is the "gold standard" treatment for early breast cancer [2] and adjuvant chemotherapy (AC) has been proved to have a significant survival benefit [3]. Although the appropriate time interval from surgery to the start of AC has not been defined, many studies demonstrated that shorter time interval was associated with better survival outcomes [4][5][6][7]. A more recent meta-analysis reported that a 4-week increase in time to initiation of AC led to a significant increase in the risk of death [8]. The initiation of AC was regularly suggested within 8 to 12 weeks after surgery [9].
While worse survival outcome from delay of adjuvant chemotherapy (DAC) has been well established, the risk factors for DAC remain unknown. Since the risk factors could not be evaluated by randomized controlled trails, evidence from numerous observational studies demonstrated that the risk factors associated with DAC included demographics, clinical characteristics, pathologic characteristics and surgical approaches[5-7, [10][11][12][13][14][15][16][17][18]. However, their impact on DAC remain inconsistent.
Due to a lack of understanding of the risk factors, we therefore conducted this systematic review and meta-analysis to identify the impact of risk factors on DAC.

Search strategy
A systematic review was conducted to identify all studies concerning the risk factors for DAC in non-metastatic breast cancer patients by searching PubMed, Embase, Chinese National Knowledge, and Wanfang Database from inception up to July 2016. Two investigators (XFH and BCZ) independently carried out the search using the following keywords simultaneously: (1) breast cancer or breast carcinoma or breast neoplasm or breast tumor; (2) adjuvant treatment or adjuvant chemotherapy; (3) delay chemotherapy or delayed chemotherapy. The reference lists of the selected articles were also reviewed for additional relevant studies.

Eligibility criteria
The inclusion criteria were as follow: the time interval between surgery and administration of AC was defined; at least one risk factor concerning DAC was investigated; odds ratio (OR) or risk ratio (RR), and associated 95% confidence intervals (CI) were available or could be calculated from the original articles. Only full-report in English was included. For duplicated cases, the most comprehensive one was eligible for inclusion. Articles were excluded if they did not meet the above criteria, or the information provided was insufficient for the outcome data extraction or quality assessment.

Data extraction
All the searched articles were independently reviewed by two investigators (XFH and BCZ). After reading the titles and abstracts, the full texts were retrieved for those potentially included articles to achieve further assessment for inclusion. Discordance in selection was solved through discussion. For the included studies, following data were extracted: author details, year of publication, data source if available, study location, sample size, age of participants, TNM stage, AC regimens if available, cut-off categorical value of time interval, any information about quality assessment under the guideline of the Newcastle-Ottawa Scale, any risk factor investigated, OR, RR and associated 95% CIs. The accuracy of extracted data was ultimately confirmed by a third investigator (FY).

Statistical analysis
Data were combined and analyzed when the risk factor was adequately considered by at least 3 studies. Because all the included studies were observational, multivariate estimates were preferentially used. If not available, univariate estimates were extracted. When the OR, RR and associated 95%CIs were not present in the original article, we calculated OR by assessing the total number of events and total number of patients in each group. The 8-week delay was determined as the cut-off time point. For studies having different time points, the closest one to the 8-week was used. We measured the inter-study heterogeneity by using I 2 statistic. Substantial heterogeneity was defined if an I 2 value exceeded 50%. Forest plots were carried out to estimate the pooled ORs using the random-effects model when I 2 value exceeded 50%, or the fixedeffects model when I 2 value not exceeding 50%. Meta-regression analysis was performed to assess the impact of year of publication and sample size on the effect on the inter-study heterogeneity. The publication bias was assessed by Egger's test. A two-tailed p-value < 0.05 was considered statistically significant. All the statistical analyses were conducted by Stata software (Stata SE 12.0). This systematic review and meta-analysis was performed under the guidelines of MOOSE [19].

Study selection
The search and selection process for eligible studies was shown in Fig 1. A total of 760 potentially relevant articles were identified, and 3 additional articles were included by manually screening the reference lists. 152 duplicates were found and removed. After reading the titles and abstracts, 569 irrelevant studies were excluded and the remaining 42 articles were reviewed in full text. Of these, 30 studies were excluded because of various reasons. Ultimately, a total of 12 articles were included for meta-analysis after detailed assessments [5-7, 10-18].

Quality assessment
To assess the quality of the observational studies, selection of participants, study comparability, and ascertainment of exposure were examined for all the included studies based on the Newcastle-Ottawa Scale [20] (shown in Table 2). A maximum of 9 starts could be obtained as the highest quality. The scores assessed for the eligible studies were ranged from 6 to 9, all of which were identified as very good or good in quality [21].

Meta-regression analysis and publication bias assessment
Meta-regression analysis suggested that year of publication and sample size did not have a significant impact on the heterogeneity across studies for each factor. Egger's test demonstrated that no potential publication bias existed among the included studies for various factors (shown in Table 3).

Discussion
In this meta-analysis, data on 186982 non-metastatic breast cancer patients from 12 studies were analyzed in characterizing the risk factors related to DAC. Combined results demonstrated that black race, rural residents and receiving mastectomy had significantly higher likelihood of experiencing DAC, while married patients were at lower risk. To the best of our knowledge, this is the first systematic review and meta-analysis evaluating the previously reported risk factors associated with DAC. Results from 9 subset studies of our meta-analysis suggested that black race was associated with an 18% increased risk of DAC compared with white race, which was consistent with the conclusions of previous studies [22,23]. However, the pooled result should be interpreted cautiously because the magnitude of race disparity on DAC was quite modest (18%) and high heterogeneity of 66.7% was observed across studies. African American women were the major component of black race in our study. The reasons for them to have a higher risk of DAC might result from following aspects: low education level, disadvantaged socioeconomic status (SES), unavailability of transportation and a lack of insurance [24][25][26][27]. Since the disparity of SES between black and white race would affect their decision on the initiation of AC after surgery [28], hence, we further divided these 9 studies into two groups: SES unknown between black and white race (U-SES), and lower SES for black race than white race (L-SES). The metaanalysis for these two groups (shown in S1 Fig) demonstrated that black race in L-SES group had a 35% increased risk of DAC, which was higher than the pooled result (18% increased risk) of the 9 studies, while there was no significant difference in U-SES group. This could partially explain that the lower SES of black race might push them to start AC administration later than white race. More work is warranted to further address this issue. In addition, our combined result from 4 studies demonstrated that mastectomy was associated with an 83% increased risk of DAC compared with BCS. Because the extent of mastectomy is larger than that of BCS, it is more likely for patients receiving mastectomy to suffer greater complications, including surgical site infections, wound dehiscence and skin flap necrosis [29,30], which could result in a longer recovery period and so delaying AC administration. A recent meta-analysis suggested that mastectomy with immediate breast reconstruction did not necessarily delay the initiation of AC compared with mastectomy only [31]. However, our study did not analyze the effect of mastectomy with immediate breast reconstruction on DAC, because no sufficient data could be extracted from the included studies. Therefore, more future studies evaluating mastectomy with immediate breast reconstruction and BCS on impact of DAC are warranted to further address this issue. Besides, three studies of the current meta-analysis documented DAC in rural residents, which was consistent with the previously reported studies and the reasons has been well interpreted that rural residents had less access to comprehensive hospitals and difficult transportation to the long-distant qualified hospitals [14,24]. Otherwise, we also observed that married patients were 42% less likely to delay the AC than single patients, since married patients usually gain more support from family members to accept clinician's recommendation and start treatment [32,33]. It is noted that several risk factors evaluated in our meta-analysis did not have significant association with DAC as mentioned in the results, which might be attributed to few studies included and inconsistent findings across included studies.
The greatest strengths of the current study are the large sample sizes of over 180000 nonmetastatic breast cancer patients and wide range of evaluated risk factors. The study indicated that black race, receiving mastectomy, rural residents and single status were significantly associated with DAC, which could be helpful for clinicians to identify the specific population groups and to start AC early. Furthermore, our work would promote the health system to pay extra attention to improve the medical conditions for patients at increased risk of delay of treatment. Of note, our meta-analysis did not focus on the survival outcomes caused by DAC. One reason is that we could not extract sufficient data from the eligible studies, since there were only 4 included studies referring to the survival outcomes. Another reason is that many previous studies and meta-analysis have demonstrated that longer time interval was associated with worse survival outcomes. Nevertheless, we did not deny that in some cases, DAC was not associated with increased risk of mortality, such as in a cohort of postmenopausal, ER-positive breast cancer patients following adjuvant endocrine therapy [34,35].
Several potential limitations of our meta-analysis should be considered. First, data were extracted from observational studies, so the inherent potential bias caused by unmeasured and uncontrolled confounders were inevitable. Second, high heterogeneity across studies was identified, although meta-regression analysis was performed to estimate the impact of year of publication and sample size and no statistically significant result was found. Thus, the interpretation of our results should be with caution. Besides, the cutoff time point of DAC was not uniform among the eligible studies, ranging from 45 to 90days, which might probably result in variability across studies and so could distort our findings. In conclusion, our meta-analysis of the current literature demonstrated that black race, rural residents, receiving mastectomy and single status led to significantly increased risk of experiencing DAC in non-metastatic breast cancer patients. Identification of these factors could be helpful for personalized treatment planning.
Supporting information S1 Fig. Forrest plots of