Characteristics of Breast Cancer Patients in Northeastern Iran: A Retrospective Cohort Study

Abstract

Background

Breast cancer (BC) is a complex disease characterized by various demographic and clinicopathological factors. Understanding the prevalence and interrelationships of these factors across regions is essential for clarifying disease progression, improving prevention and early detection strategies.

Design

This retrospective cohort study included 4043 female breast cancer patients treated at a radiotherapy & oncology center in northeastern Iran. Due to complete-case analysis, sample sizes varied across analyses.

Objectives

To describe the frequency of demographic and selected clinicopathological characteristics and to investigate their interrelationships.

Methods

Data preprocessing was performed to ensure accurate and effective presentation of the results. Univariate log-linear analyses were first conducted to assess crude associations between variables. Variables of interest were then entered into a multivariable log-linear model adjusted for potential confounders, including age at diagnosis, estrogen receptor (ER) status, and progesterone receptor (PR) status.

Results

Patients residing in rural areas showed a significant associationwith a time to presentation longer than five months (OR = 1.48, 95% CI = 1.12–1.96), additionally educated patients significantly associated with shorter time to presentation (OR = 0.69, 95% CI = 0.58–0.83). No significant association was observed between age at diagnosis and time to presentation. Likewise, no association was found with various statuses for HR/HER2, or grade and time to presentation. Education, employment, place of residence and marital status showed no statistically significant associations with clinicopathological characteristics. Any observed association in unadjusted models could be explained by age at diagnosis, ER status, or time to presentation. Multiple tumors were associated with higher odds of ER/PR positivity (OR = 1.42, 95%CI = 1.01-1.99) and were negatively associated with age 40–60 compared to those under 40 (OR = 0.64, 95%CI = 0.44-0.93). Tumor laterality showed no statistically significant relationships with hormone receptor (HR) status, histological grade, employment, education, place of residence, or marital status.

Conclusion

This study identified associations between several variables. However, the observed associations may be influenced by unmeasured factors. Further studies are needed to clarify the underlying mechanisms driving these relationships.

Plain Language Summary

Breast cancer is influenced by many personal and medical factors, and these factors may differ between regions. Understanding these differences can help improve early diagnosis, treatment, and health planning. In this study, we examined information from about 4,034 women diagnosed with breast cancer who were treated at a medical center in northeastern Iran. We looked at basic characteristics such as age at diagnosis, education level, employment status, marital status, and whether patients lived in urban or rural areas. We also examined medical features of the cancer, including tumor histologic type, hormone receptor status, tumor grade, laterality, and whether patients had more than one tumor. Statistical analyses were used to explore how these characteristics were related. Our findings showed that women with lower education levels and those living in rural areas were more likely to experience longer delays before starting treatment. Women living in rural areas were also less likely to be over 60 years old. However, rural or urban residence was not linked to estrogen or progesterone receptor status or tumor grade. Older women were more likely to have estrogen receptor–positive tumors and tumors that were well or moderately differentiated. Women over 40 years of age were less likely to have more aggressive breast cancer subtypes compared with younger women. Patients with multiple tumors were more likely to have hormone receptor–positive disease. No clear relationships were found between tumor laterality and patient characteristics, or between hormone receptor status and education, employment, or marital status. Overall, this study highlights important patterns in breast cancer characteristics in northeastern Iran. Some findings may be influenced by factors not measured in this study, and further research is needed to better understand these relationships and improve cancer care.

1. Introduction

Breast cancer (BC) is one of the most prevalent and potentially life-threatening diseases, affecting millions of women worldwide.^1-5 The number of new breast cancer cases is estimated to exceed 3 million annually, with more than 1 million deaths expected each year. ⁶ Additionally, BC ranks as the fifth leading cause of cancer-related deaths globally and is the most common cause of cancer mortality among women in less developed countries. ⁷ In Iran, breast, colorectal, and stomach cancers were among the most common cancers in 2016 and were estimated to remain among the most frequently diagnosed cancers nationally in 2025. ⁸ Although, the age-standardized incidence rate (ASR) of breast cancer in Iran remains lower than in other European and neighboring countries such as Qatar, Turkey, and Iraq, ⁹ studies show the trend is rising.^10-13

Recently, advances in breast cancer diagnosis and prognosis, particularly the integration of Artificial Intel- ligence (AI), have made it easier to detect patients at early stages and have significantly enhanced diagnostic accuracy. However, it is recognized that these methods should be culturally acceptable to ensure effective participation and adherence.^14-17 As a result, studies show the survival rate has been improved over years, ¹⁸ nonetheless, challenges and disparities in outcomes still remain a concern.

The etiology of breast cancer involves a combination of environmental, nutritional, hormonal, and hereditary factors that contribute to disease development. Exploring how these factors interact provide insight into underlying mechanisms and help improve prevention and early-detection strategies.^19-21 The demographics and cancer- specific characteristics of breast cancer patients have been examined in studies worldwide. Sisti et al. ²² conducted a descriptive analysis of breast cancer in women using data from the national cancer database in the United States for the period 2004-2015. Soleimani et al. ²³ utilized data from the Iranian national population-based cancer registry, as well as environmental data at the provincial level in Iran from 2014 to 2018. They underscore the significant of predictors such as marriage, tertiary education attainment rate, physician-to-population ratio, and air pollution. Akbari et al. ²⁴ investigated the characteristics of 3010 women with breast cancer between 1998 and 2014. Pourriahi et al. ²⁵ examined the demographic and clinical characteristics of 1476 patients in Iran. While these studies have investigated the frequency of breast cancer characteristics in patients across other regions of the country, to our knowledge, no study has thoroughly examined breast cancer patients in the northeast of Iran. The present study addresses this gap by analyzing data from a radiotherapy & oncology center in northeastern Iran between March 21, 2019, and March 21, 2024. Unlike previous studies, including the recent national Iranian cohort, ¹⁸ which focused primarily on survival outcomes, the present study provides detailed insight into regional epidemiological patterns with statistical rigor, focusing on associations among demographic, socioeconomic, clinical, and pathological variables. Specifically, this study complements prior research by: (1) providing regional granularity for northeastern Iran, (2) examining clinicopathological associations rather than only descriptive statistics (frequencies and percentages) or survival outcomes, and (3) focusing on recent years (2019–2024) to account for changes in clinical practices and socioeconomic factors, such as education and access to healthcare. In this study, findings that are well established in the literature receive less focus, while more attention is devoted to those that are conflicting or understudied. This mainly leads to the following research questions:

• 1. What are the frequency and distribution of demographic, clinical, and pathological characteristics among female breast cancer patients in northeastern Iran, and how do these compare with those reported in global populations, as well as in other regions of Iran?

• 2. Are socioeconomic factors (education, employment, marital status, and place of residence) associated with clinical outcomes (tumor grade, receptor status, number of tumors, and laterality)?

• 3. Are socioeconomic factors (including education, employment, marital status, and place of residence) and clinicopathological characteristics (tumor grade, receptor status) associated with delayed presentation?

2. Materials and Methods

2.1. Dataset Description

This study was carried out on breast cancer patients, admitted over a five-year period from 21-03-2019 to 21-03- 2024, at the Reza Radiotherapy and Oncology Center (RROC) in Mashhad, Razavi Khorasan, Iran. RROC is recognized as one of the leading facilities for cancer detection and treatment in the region, while offering various cancer treatment modalities including chemotherapy and others, its primary focus lies in providing radiation therapy as an integral component of cancer care.

When a person is diagnosed with breast cancer, treatment typically begins with either surgery followed by chemotherapy, known as adjuvant treatment, or chemotherapy given before surgery to shrink the tumor, known as neoadjuvant treatment. This is often followed by additional therapies such as radiation or hormone therapy. The majority of patients come to RROC for radiation treatment, and in many cases, patients may receive the remaining components of their treatment at another medical facility. However, the pathological and clinical information of patients is recorded at the time of admission.In cases of recurrence, patients requiring further radiotherapy return to the center for additional treatment. In this study, admission records were utilized to gather relevant variables. Demographic, family history and reproductive data were recorded based on patient self-reports, while information on diagnosis, histology of tumor, surgery, and patient’s condition was documented using clinical reports. Note that selecting patients from a center whose primary function is to provide radiotherapy introduces bias into the dataset. Such a center is likely to be enriched for cases with higher-stage or more aggressive breast cancer, patients who receive only surgery without radiotherapy are rarely treated at this facility.

The initial dataset included 5340 patients and approximately 100 variables. The data cleaning process began by integrating information related to patient diagnosis, symptoms, surgery, family history, reproductive factors and treatment based on shared document codes across all files. Under clinician guidance, fundamental cleaning procedures were implemented. First, a thorough analysis of variables was performed, considering their names, meanings, types, and values to retain only distinct and pertinent information. Variables with singular values were omitted, after extracting all distinct information, duplicate rows were removed based on document numbers. Any detected inconsistencies among unique variable values were carefully addressed. Additionally, values that fall below the 25th percentile minus 1.5 times the interquartile range (IQR) or above the 75th percentile plus 1.5 times the IQR were considered outliers. These outliers were treated as missing values in subsequent analyses. Dates of biopsy, surgery, radiotherapy, birth, admission, and recurrence were available in the dataset. Some information, such as age at diagnosis, was inferred from these dates. The number of tumors was determined based on the number of reported pathological tumor sizes. Analysis was restricted to patients whose biopsy and surgery dates fell within the study timeframe, to minimize uncertainty, patients with unknown surgery and biopsy dates were also excluded, even if other dates were within the study timeframe. In some cases, radiotherapy treatment dates may have fallen outside the study timeframe. After these refinements, the final dataset comprised information from 4551 patients with 49 variables. Due to space limitations, this study focuses on the relationships between selected variables including age, education, employment status, place of residence, and marital status, time to presentation, grade, laterality, number of tumors, estrogen receptor (ER) status, ER intensity, progesterone receptor (PR) status, PR intensity, human epidermal growth factor receptor 2 (HER2) status, molecular subtype, histology, and the examination of other variables is reserved for future research. Age and time to presentation are continuous variables, while other variables are categorical. Summary statistics (mean (SD) and median (IQR)) are reported for continuous variables, and frequencies with percentages for categorical variables. However, for statistical and descriptive analysis, continuous variables were categorized.

2.2. Descriptive Analysis

During the study period, a total of 4551 BC patients received treatment at RROC. Figure 1 illustrates the number of patients admitted each year throughout this period.

Figure 1.

A total of 4551 patients received radiotherapy treatment at RROC between March 21, 2019, and March 21, 2024. Patients with recurrence were counted only once for their initial breast cancer treatment

We excluded patients whose initial diagnosis occurred outside the study framework (221 (4.8%)) or who presented with metastatic disease at diagnosis (n = 261(5.7%)). This resulted in a total of 4069 patients with breast cancer. Of these, 4034 (99.1%) were female and 35 (0.9%) were male. Male patients were excluded from subsequent analyses (Figure A1). The majority of patients were female (n = 4034), 3923 of them were from Iran and around 115 patients from other countries such as Afghanistan, Iraq and Turkmenistan, Information on country of birth was missing for 5 patients.

2.2.1. Demographic Information

The age distribution of female patients (n = 4034) is shown in Figure 2. The highest number of cancer diagnoses occurred in women aged 40–60 years (n = 2464, 61.08%), followed by those over 60 years (n = 874, 21.66%). Women under 40 years had a lower frequency of diagnoses compared to older age groups (n = 696, 17.25%).

Figure 2.

Age distribution for female patients (n = 4034), the mean (standard deviation (SD)) age was 50.68 (11.39) years, with a range of 16.9 to 90.4 years old. The median age [IQR] was 49.5 [42.3–58.4] years

Table 1 presents the demographic characteristics of the study population. Patients categorized as Married/Divorced/Widowed (MDW) and unemployed represented the largest groups in the dataset. Similarly, most patients had urban place of residence. Educational attainment was approximately evenly distributed, with nearly half of the patients having qualifications below a diploma and the other half having a diploma or higher level of education.

Table 1.

Demographic Characteristics of Patients (n = 4034).

Variable	Frequencies (%) a	Description
Education
Below diploma	2008 (49.77%)	Diploma is equivalent to 12 years of study
Diploma and above	2006 (49.72%)
Missing	20 (0.49%)
Employment status
Unemployed	3114 (77.19%)	Laid off, student, retired are considered as employed.
Employed	856 (21.21%)
Missing	64 (1.58%)
Place of residence
Urban	3617 (89.66%)	Urban areas include cities with a population above 2000 people.
Rural	410 (10.16%)
Missing	7 (0.17%)
Marital Status
Married/Divorced/Widowed	3852 (95.5%)	​
Single	179 (4.5%)
Missing	3 (0.07%)

Most patients were unemployed, lived in urban areas, and were married. patients’ educational attainment was roughly evenly split between below diploma and diploma and above level

^aThe percentages were calculated out of the initial number of patients (n = 4034).

2.2.2. Time to Presentation

We investigated variation in the length of patient presentation delay, defined as the time between a patient’s first awareness of breast symptoms (e.g., palpable mass, skin changes, nipple discharge, among others) and their first medical contact with a physician at any healthcare facilities. The mean (SD) was 2.81 (3.24) months, with a range of 0.1 to 13 months. The median (IQR) length of time to presentation across patients was 1 month (1–4 months). Information on time to presentation was available for 3163/4034 patients. Time to presentation was categorized into three groups, those who presented within 1 month, those who presented within 1-5 months, and those who presented at 5 months or more. The majority of patients, 1665/3163 (52.63%) sought medical consultation within 1 month or less, 837/3163 (26.46%) had a time of 1-5 months, and 661/3163 (20.89%) had a time exceeding 5 months. Time to presentation data were missing for 871/4034 (21.59%) patients.

2.2.3. Clinicopathological Information

Information on number of tumor, laterality (left vs. right breast), histological grade, and histology is summarized in Table 2. Most patients had a single breast tumor, with a slight predominance of left-sided disease. Histological grade was classified as well differentiated (Grade I), moderately differentiated (Grade II), or poorly differentiated (Grade III), with Grade II tumors being the most common. Tumor histology comprised invasive ductal carcinoma, lobular carcinoma, mixed histological types, carcinoma in situ, and other less common subtypes, with invasive ductal carcinoma being the most prevalent.

Table 2.

Clinicopathological Characteristics of the Study Population.

Variable	Frequencies (%) a	Description
Number of tumors
1	2894 (71.74%)	The total number of tumor lesions identified in the breast, regardless of whether they were associated with multifocal (MF) or multicentric (MC) disease.
≥2	195 (4.83%)
Missing	945 (23.43%)
Laterality
Left	2054 (50.91%)	​
Right	1879 (46.57%)
Both sides	15 (0.37%)
Missing	86 (2.13%)
Grade
Poorly differentiated	1301 (32.25%)	Histological tumor grade was evaluated using the Nottingham grading system.
Moderately differentiated	2020 (50.07%)
Well differentiated	318 (7.88%)
Missing	395 (9.79%)
Histologic type
Ductal	3541 (87.77%)	Tumor histology was classified according to the International Classification of Diseases for Oncology, third edition (ICD-O-3).
Lobular	183 (4.53%)
Mixed	141 (3.49%)
In situ	104 (2.57%)
Other	32 (0.79%)
Missing	33 (0.81%)

The majority of patients had a single tumor, left-sided disease, moderately differentiated tumors, and ductal invasive carcinoma histology

^aThe percentages were calculated out of the initial number of patients (n = 4034).

The distributions of ER and PR status, their intensity, HER2 status, and molecular subtype are presented in Table 3. ER and PR status are reported as positive or negative, and staining intensity is categorized as weak, moderate, or strong. HER2 status is reported as positive or negative, and molecular subtype is classified as luminal A, luminal B, HER2-enriched, or triple-negative based on the expression of ER, PR, and HER2.

Table 3.

Distribution of ER and PR Status and Intensity, HER2 Status, and Breast Cancer Molecular Subtypes.

Variable	Frequencies (%) a	Description
ER status
Positive	2043 (50.45%)	ER or PR status was defined as positive when more than 1% tumor cells were positive for either marker using the Allred scoring system.
Negative	1232 (30.59%)
Missing	759 (18.81%)
ER Intensity
Strong	1544 (75.6%)	ER intensities were calculated only for ER-positive patients (n = 2043).
Moderate	396 (19.4%)
Weak	103 (5.0%)
Missing	0 (0%)
PR status
Positive	1751 (43.35%)	​
Negative	1524 (37.82%)
Missing	759 (18.8%)
PR Intensity
Strong	1200 (68.5%)	PR intensities were calculated only for PR-positive patients (n = 1751).
Moderate	449 (25.6%)
Weak	102 (5.8%)
Missing	0 (0%)
HER2 status
Negative	2330 (57.75%)	HER2 was positive when tumor IHC b score was 3+, or negative when IHC score was 1+, or for an IHC score of 2+, FISH/CISH b score was considered.
Positive	757 (18.76%)
Missing value c	947 (23.47%)
Molecular subtype
Luminal A	1374 (34.06%)	Luminal A: Either ER or PR strongly positive, HER2 negative. Luminal B: Either ER or PR moderately/weakly positive, HER2 negative, or -either ER or PR positive. HER2-enriched: ER, PR negative, and HER2 positive. Triple-negative: ER, PR and HER2 negative.
Luminal B	717 (17.77%)
HER2-enriched	396 (9.81%)
Triple-negative (basal-like)	788 (19.53%)
Missing value	759 (18.81%)

ER and PR positivity were more common than negativity, with strong intensity predominating among receptor-positive tumors. Most tumors were HER2-negative, and luminal A was the most frequent molecular subtype, followed by luminal B.

^aThe percentages were calculated out of the initial number of patients (n = 4034).

^bIHC: Immunohistochemistry, FISH: Fluorescence In Situ Hybridization, CISH: Chromogenic In Situ Hybridization.

^cAmong the 947 patients, 100 had equivocal HER2 results with missing FISH/CISH confirmation.

The high percentages of missing data (MD) among several clinicopathological variables raised concerns about potential bias and loss of information in subsequent analyses. In the following section, we thoroughly explain the underlying missing data mechanisms, the methods used to address missingness, and the derivation of the analytical sample for each analysis.

2.3. Statistical Analysis

In this study, we first applied pairwise log-linear models to further investigate the associations among the categorical variables and quantify the strength of these relationships. Pairwise log-linear models were selected instead of logistic regression because our primary objective was to evaluate association patterns among categorical variables rather than to model the effects of explanatory variables on a single categorical response. This approach treats all variables symmetrically and is particularly appropriate for investigating association patterns and dependence structures among categorical variables. ²⁶ A homogeneous association model, defined as a multivariable log-linear model allowing conditional associations between all three pairs of variables, was then fitted. The models were adjusted for age at diagnosis and ER/PR status based on their established clinical relevance in breast cancer. Where appropriate, further adjustments were made for additional demographic and clinical factors, including education level, time to presentation, and employment status. We ensured that the assumptions for log-linear analysis were satisfied, namely that the expected frequencies were greater than or equal to 5 for at least 80% of the categories, and all expected frequencies were greater than 1. When these assumptions were violated, the results were reported as ’NA’. Otherwise, the results were reported as the number of observations (n), odds ratios (OR), 95% confidence intervals (CIs), and p-values. Adjusted odds ratios (AORs) were reported for the multivariable model. Before fitting the model, we calculated variance inflation factors (VIFs) to assess potential multicollinearity among highly correlated variables. ²⁷ Note that for certain variables, multivariable analysis was not feasible due to zero counts in some contingency table cells. To address this issue, we added a small constant (0.2) to the zero cells to avoid computational problems associated with empty cells. ²⁸

All analyses were conducted using Python version 3.12, with the following libraries: NumPy, SciPy, pandas, statsmodels, seaborn, and miceforest.

2.3.1. Missing Data

Tables 1-3 present the variables included in the current study along with the corresponding number of missing cases. As it can be seen, the extent of MD varied across variables. Demographic variables exhibited minimal missingness, ranging from 0.07% to 1.5%, whereas clinicopathological variables showed higher levels of missingness, ranging from 0.8% to 23.5%. The proportion of missing values for time to presentation was 21.59%. The missingness in the clinicopathological and demographic variables in our dataset appeared to reflect a combination of two mechanisms: missing completely at random (MCAR) and missing at random (MAR). Missing values in baseline demographic variables were primarily attributable to occasional human error during routine clinical data entry. Although a standardized data collection protocol was implemented to minimize missing infor mation, random operational oversights by registration staff occasionally resulted in incomplete records. As these omissions were not systematically related to patient characteristics or clinical outcomes, they were considered broadly consistent with an MCAR mechanism.

In contrast, missingness in several clinicopathological variables was more plausibly explained by a MAR mechanism. Many patients were referred to the center exclusively for radiotherapy after receiving their initial diagnosis, surgery, and pathological evaluation at external institutions. Consequently, complete baseline pathological information was often unavailable in the RROC database because these records had not been transferred at the time of data collection. Conversely, patients who received their diagnosis and subsequent treatment, including chemotherapy, at RROC had lower rates of missing data. This pattern suggests that the probability of missingness was associated with observed characteristics related to the referral pathway and treatment history, making the MAR assumption more plausible for these variables.

To handle missing data, listwise deletion (complete case analysis, CCA) was applied. This approach restricted the analysis to patients with complete data for all variables of interest, which resulted in varying sample sizes (n) across the different analyses. The main advantage of listwise deletion is its simplicity and its compatibility with a wide range of statistical methods. When data are MCAR, listwise deletion does not introduce bias, because the complete cases effectively represent a simple random sample of the original dataset; since simple random sampling is unbiased, the analysis remains valid under MCAR. However, it is well established that complete case analysis may still lead to reduced efficiency even under the MCAR assumption, as standard errors are typically larger than necessary, resulting in wider confidence intervals and potentially higher p-values. ²⁹

When data are not MCAR, the subset of cases with complete observations is not representative of the full dataset. The resulting bias and reduced precision depend on several factors: the proportion of complete cases, the structure of missingness across variables, how much the complete and incomplete cases differ from one another, and the specific population parameter being estimated. CCA is generally considered acceptable when the proportion of missing data is relatively small compared with the overall sample size, although some loss of statistical power may still occur. ³⁰ In the present study, the initial sample size was 4,034 patients, and the proportion of complete cases remained sufficiently large after applying CCA, particularly for the certain variables, which had low levels of missing data.

Additionally, we examined whether the distribution of variables differed between included and excluded cases for variables with more than 18% missingness. For the ER status analysis, only tumor histology differed signif icantly between included and excluded cases, the excluded group had a higher proportion of in situ and mixed histologic types, while the included group had a higher proportion of invasive ductal histology (Table A10). A similar pattern was observed for the PR status analysis, these results are not shown separately to avoid redundancy. For the HER2 status analysis, there are significant differences between included and excluded cases in tumor histology and hormone receptor (HR) status, excluded patients had a lower proportion of HR positivity than included patients, along with a slightly higher proportion of mixed and other histologies. Turning to Table A11, significant differences in age, tumor grade, and HR status were observed between patients included in and excluded from the time-to-presentation analysis. Included patients were younger and had more aggressive tumor features than excluded patients, with a higher proportion under age 40 and a higher proportion of poorly differentiated tumors; excluded patients were older, more often had well- or moderately differentiated tumors, and showed slightly higher rates of HR positivity. For the number-of-tumors analysis, significant differences were observed in age, education, marital status, and receptor status (ER, PR, and HER2). Excluded patients were younger and more often single than included patients, with a higher proportion under age 40, a lower proportion over age 60, lower rates of HR positivity, and higher rates of HER2 positivity. It is worth noting that this test offers no direct evidence on the validity of the MAR assumption, and has limited power when the sample of incomplete units is small. ³⁰

As we noted, when the missing data mechanism is likely to be MAR, relying only on complete cases can produce biased results. To reduce this bias, imputation methods can be used, allowing participants with incomplete data to remain in the analysis. In this study, multiple imputation (MI) ³¹ was applied to evaluate the robustness of the observed associations among variables that showed significant differences between the included and excluded participants. Specifically, these included the associations between the number of tumors and age, education, marital status, ER/PR status, and HER2 status, as well as the associations between time to presentation and age, tumor grade, and ER/PR status. The imputation model included education, job status, marital status, place of residence, number of tumors, ER status, PR status, HER2 status, tumor grade, age at diagnosis, time to presenation. Twenty imputed datasets were generated using the miceforest package (version 6.0.5) in Python, ³² which implements Multiple Imputation by Chained Equations (MICE) framework using LightGBM, a gradient boosting decision tree algorithm, ³³ to model the conditional distribution of each variable with missing values. During each iteration, every incomplete variable was imputed sequentially using the remaining variables in the imputation model as predictors. This process was repeated to generate each imputed dataset (Table A12). Log-linear models were then fitted separately to each imputed dataset. The resulting coefficient estimates and standard errors were pooled using Rubin’s rules, ³⁴ which combine within-imputation and between-imputation variability to produce valid standard errors, confidence intervals, and p-values. Results were compared against the primary complete-case analysis to evaluate the consistency and robustness of the observed associations (Tables A13 and A14).

3. Results

In this section, we investigated the relationships between variables in our dataset. Due to the listwise deletion approach applied in this study the sample size (n) fluctuated across analyses depending on the missingness profile of the variables involved. The final sample sizes for these analyses ranged from a minimum of 2364 to a maximum of 4031. The percentages were calculated using the number of non-missing values for each variable as the denominator. In addition, given the large sample size, several associations reached statistical significance despite modest effect sizes, where relevant, we note this explicitly, as odds ratios close to 1 may carry limited clinical relevance even when statistically significant.

3.1. Age at Diagnosis

The associations between age at diagnosis and other variables are presented in Table A1.

As it can be seen, no association was observed between age at diagnosis and tumor laterality (n = 3948, p-value > 0.4), the distribution of left or right breast involvement did not differ significantly across age groups, this finding is consistent with another study conducted in Egyptian breast cancer patients. ³⁵ After adjusting for ER status in the model this relation persisted (n = 3212, p-value > 0.6).

In contrast, compared with patients younger than 40 years, those aged between 40–60 years had 36% lower odds of reporting more than two tumors (n = 3088, OR = 0.64, 95% CI: 0.44–0.93, p-value ≤ 0.001). This association persisted after adjustment for ER status, indicating that it was not fully explained by differences in ER status. However, no association was observed initially between age at diagnosis over 60 and the number of tumors, after adjusting for ER status within the statistical model, the association between age at diagnosis over 60 and number of tumors became statistically significant, patients aged over 60 had 50% lower odds of presenting with multiple tumors in the multivariable analysis (n = 2549, OR = 0.50, 95% CI: 0.30–0.84, p-value ≤ 0.01) compared with patients under 40 years of age. Because the number of tumors contained a substantial proportion of missing values (23.43%), and the age distribution differed between patients included in and excluded from the complete-case analysis (Table A11), a multiple imputation analysis was performed as described in Section 2.3.1. The imputed analysis produced results that were consistent with those obtained from the complete-case analysis (Table A13). When considering poorly differentiated grade as the reference category, age at diagnosis between 40–60 were initially associated with a moderately differentiated grade (n = 3639, OR = 1.30, 95% CI: 1.07–1.56, p-value ≤ 0.01) or a well differentiated grade (n = 3639, OR = 1.66, 95% CI: 1.13–2.44, p-value ≤ 0.01). After adjusting ER/PR, the relation between age at diagnosis 40–60 and grade (both level) were not significant (n = 3003, p- value > 0.08), suggesting that hormone receptor status may confound the relationship between age and tumor grade. Additionally, patients with age at diagnosis over 60 were associated with a moderately differentiated grade tumors (n = 3639, OR = 1.60, 95% CI: 1.28–2.01, p-value ≤ 0.001) or a well differentiated grade (n = 3639, OR = 2.90, 95% CI: 1.90–4.42, p-value ≤ 0.001), indicating compared with patients younger than 40 years, those over 60 years had 60% higher odds of having moderately differentiated tumors and almost threefold higher odds of having well differentiated tumors. The larger effect size for well differentiated tumors suggests that older patients were more likely to present with less aggressive tumor differentiation. Unlike the 40–60 group, this association for patients over 60 remained statistically significant after adjusting for ER/PR status.

To investigate association between age at diagnosis and histology, we repeated the main analysis by considering various level of histology as the reference category. With ductal histology as the reference category, it can be seen that patients with age at diagnosis 40–60 and lobular histology initially showed an association (n = 4001, OR = 1.81, 95% CI: 1.09–2.99, p-value ≤ 0.05). Patients with age at diagnosis over 60 and lobular histology showed a significant association (n = 4001, OR = 2.16, 95% CI: 1.25–3.76, p-value ≤ 0.01). After adjustment for ER/PR status, the association between age greater than 60 years and histology remained statistically significant, whereas the association for patients aged 40–60 years was attenuated and no longer reached statistical significance (n = 3252, p-valu > 0.8). These findings indicated that patients aged over 60 years had higher odds of lobular histology relative to ductal histology compared with patients diagnosed before 40 years of age. Our results were consistent with the study by, ³⁶ which reported that invasive lobular carcinoma was associated with a higher age at diagnosis. No statistically significant association between age and in situ tumors was observed when compared with ductal tumors.

Using in situ histology as the reference category, lobular histology was significantly associated with age at diagnosis (OR = 2.32, 95% CI: 1.16–4.61, p-value = 0.01 for age 40–60, and OR = 3.68, 95% CI: 1.61–8.43, p-value ≤ 0.01) for age ≥ 60), indicating that patients aged 40 years and older had higher odds of having invasive lobular tumors relative to in situ tumors than patients younger than 40 years. However, the wide confidence intervals, particularly for patients aged over 60 years, suggested some uncertainty on odds ratios despite the statistically significant associations.

The observed associations with age and hormone receptor status were consistent with previous reports indicating that younger women are more likely to present with hormone receptor–negative breast cancer subtypes.^37-40 Due to space limitations and the well-established nature of these results, they are not reproduced here. Refer to Table A1 for the associations found between age and HR status, thier intensity, HER2 status, and molecular subtype in our analysis.

3.2. Education

As indicated in the Table A2, patients with diploma and above education were associated with age at diagnosis between 40-60 (n = 4014, OR = 0.35, 95% CI: 0.29–0.42, p-value ≤ 0.001), and age over 60 (n = 4014, OR = 0.15, 95% CI: 0.11–0.18, p-value ≤ 0.001). These findings indicate that compared with patients younger than 40 years, those aged 40–60 years had 65% lower odds of holding a diploma or higher qualification, while patients older than 60 years had 85% lower odds of having attained at least a diploma. This pattern is consistent with population-level educational trends, as younger generations generally have greater access to formal education and therefore tend to have higher educational attainment.

A significant association was observed between education and place of residence. Patients with a diploma or higher level of education had 87% lower odds of residing in rural areas than those with lower educational attainment (n = 4010, OR = 0.13, 95% CI: 0.10–0.17, p-value ≤ 0.001). Furthermore, compared with patients whose educational attainment was below a diploma, those with a diploma or higher had 22.87 times higher odds of being employed (n = 3955, OR = 22.87, 95% CI: 17.38–30.08, p-value ≤ 0.001). In addition, they had 2.61 times higher odds of being single than patients with education below a diploma (n = 4012, OR = 2.61, 95% CI: 1.86–3.64, p-value ≤ 0.001). Patients with a diploma or higher level of education had lower odds of having a moderately differentiated tumor compared with a poorly differentiated tumor than those whose educational attainment was below a diploma (n = 3620, OR = 0.82, 95% CI: 0.72–0.94, p-value ≤ 0.001). After adjustment for age at diagnosis, the association between education and tumor grade was no longer statistically significant (n = 3620, p-value = 0.23), suggesting that differences in age distribution may have contributed to the observed association. It is worth noting that the initial unadjusted association had a small effect size, and its statistical significance may have reflected the robust statistical power afforded by our large sample size.

Considering invasive ductal histology as the reference category, diploma and above education level showed a significant association with mixed histology (n = 3981, OR = 1.57, 95% CI: 1.10–2.24, p-value ≤ 0.01). Patients with mixed histology had 57% increased odds of having a diploma or higher level of education. Diploma and above and in situ histology also showed a significant association (n = 3981, OR = 1.72, 95% CI: 1.15–2.57, p- value ≤ 0.01). However, the association between education and in situ histology lost statistical significance after adjusting for ER/PR status as a confounder in the multivariable model (n = 3233, p-value = 0.3). No statistically significant association was observed between education level and lobular histology compared with invasive ductal carcinoma (n = 3981, p = 0.4).

Using the in situ histology category as the reference, lobular histology showed a significant negative association with higher education level (n = 3981, OR = 0.45, 95% CI: 0.28–0.74, p-value ≤ 0.01). Specifically, patients with lobular histology had 55% lower odds of having a diploma or higher level of education than those with in situ histology in unadjusted model, after adjustment for age at diagnosis, the previously observed association was no longer statistically significant (n = 3981, p-value > 0.56).

There was a positive association between diploma-and-above education and moderate ER intensity (n = 2031, OR = 1.51, 95% CI: 1.18-1.87, p-value ≤ 0.001), initially. After adjusting age at diagnosis, the relationship is not significant (n = 1975, p-value > 0.07).

In contrast with, ⁴¹ we did not find any association between educational level and ER status (n = 3256, p- value = 0.76), and PR status (n = 3256, p-value = 0.89) in our dataset. Similarly, no association was observed between educational level and HER2 status (n = 3067, p-value > 0.17), or molecular subtype (n = 3256, p-value > 0.10), consistent with findings reported by. ⁴² Overall, educational level was not independently associated with tumor biology (grade, receptor status, or molecular subtype) in breast cancer. Any unadjusted associations with educational level were explained by confounding with age at diagnosis or ER status.

3.3. Employment Status

As indicated in the Table A3, using unemployed patients as the reference category, age at diagnosis was not significantly associated with employment status in the univariable analysis (n=3970, p-value > 0.9). However, after adjusting for education in the multivariable model, the association became statistically significant. Compared with patients younger than 40 years, those aged 40–60 years had higher odds of being employed rather than unemployed (n = 3955, OR = 2.25, 95% CI: 1.71–2.70, p-value ≤ 0.001). Similarly, patients older than 60 years had higher odds of being classified in the employed category (n = 3955, OR = 4.35, 95% CI: 3.21–5.85, p-value ≤ 0.001). These findings are of interest because we initially expected age and employment status to show a similar pattern to age and education. In this cohort, older patients were associated with lower odds of having a diploma or above, and younger patients generally had higher educational attainment. Given this, we expected younger patients to also show higher odds of being classified in the employed category. However, after adjusting for education, older patients were associated with higher odds of being employed compared with younger patients, indicating that once education was held constant, the association between age and employment status reversed direction.

The odds of being employed were 86% lower among patients residing in rural areas compared with urban-residing patients (n = 3967, OR = 0.14, 95% CI: 0.09–0.24, p-value ≤ 0.001). After adjusted education (n = 3952), ER status (n = 3221) this association remained significant.

Employment status showed a significant association with marital status, where the odds of being single were higher among employed patients (n = 3968, OR = 2.37, 95% CI: 1.72–3.26, p-value ≤ 0.001). Employed patients had 137% higher odds of being single than unemployed patients.

Considering invasive ductal histology as reference category, patients with employment status as employed showed a significant association with in situ histology (n = 3938, OR = 1.59, 95% CI: 1.04–2.44, p-value ≤ 0.05), this effect disappeared with considering education as cofounding (n = 3923, p-value = 0.4). The association between employment status and tumor histology also lost statistical significance after adjusting for ER/PR status (n = 3202, p-value = 0.3). The VIF for education and employment status was 1.26, indicating no evidence of multicollinearity.

In the crude model, employed patients showed a negative association with luminal B compared to luminal A molecular type (n = 3224, OR = 0.76, 95% CI: (0.61–0.96), p-value ≤ 0.05). However, this effect size was small, and we assume that this may be related to the large sample size, which can detect differences that are not necessarily clinically important. Additionally, after adjusting the model for time to presentation, this association was no longer statistically significant (n = 2657, p-value 0.08), suggesting that employed patients did not differ in tumor subtype from unemployed patients after accounting for variations in presentation time.

No statistically significant associations were observed between employment status and tumor-related clinicopathological characteristics. Specifically, employment status was not significantly associated with the number of tumors (n = 3044, p-value = 0.43), tumor laterality (n = 3888, p-value = 0.72), histological grade (n = 3581, p- value ≥ 0.32), ER status (n =3224, p-value = 0.78), PR status (n =3224, p-value = 0.8), HER2 status (n =3039, p-value = 0.34).

3.4. Place of Residence

As indicated in Table A4, rural place of residence showed a significant association with age at diagnosis, where the odds of being over 60 years old were 61% lower among patients residing in rural areas compared to those residing in urban areas (n = 4027, OR = 0.39, 95% CI: 0.26–0.56, p-value ≤ 0.001).

Rural residence was also significantly associated with positive HER2 status (n = 3082, OR = 1.35, 95% CI: 1.01– 1.70, p-value ≤ 0.05). However, this association did not remain statistically significant after adjustment for age at diagnosis in the multivariable analysis (n = 3082, OR = 1.29, 95%CI : 1.00-1.68, p-value = 0.052).

No statistically significant associations were observed between place of residence and tumor grade (n = 3634, p-value ≥ 0.1), tumor number (n = 3083, p-value = 0.3), laterality (n = 3943, p-value = 0.3), ER/PR status (n = 3270, p-value = 0.5), molecular subtype (n = 3270, p-value ≥ 0.1), or tumor Histologic types (n = 3890, p-value ≥ 0.07).

3.5. Marital Status

Single marital status was significantly associated with age at diagnosis, where the odds of being single were 69% lower among patients aged 40–60 (n = 4031, OR = 0.31, 95% CI: 0.23–0.43, p-value ≤ 0.001), and 90% lower among those aged over 60 (n = 4031, OR = 0.10, 95% CI: 0.05–0.19, p-value ≤ 0.001) than those diagnosed under the age of 40.

Initially, single marital status showed a negative significant association with well differentiated grade (n = 3637, OR = 0.26, 95% CI: 0.09–0.74, p-value ≤ 0.05), however, after adjusting for age at diagnosis, the relation between marital status and grade lost statistical significance (n = 3273, p-value = 0.3), which may reflect the fact that single patients in our cohort were more likely to be under 40, and patients under 40 were more likely to have high-grade tumors, partly explaining the higher frequency of high-grade tumors among single patients prior to adjustment.

Considering invasive ductal histology as reference category, patients with marital status as single showed a significant association with in situ histology (n = 3998, OR = 2.28, 95% CI: 1.17–4.48, p-value ≤ 0.05). However, the association between marital status and histology could not be examined in an adjusted model incorporating age at diagnosis and ER/PR status due to data sparsity.

No statistically significant associations were observed between marital status and tumor tumor number (n = 3088, p-value = 0.1), laterality (n = 3946, p-value = 0.4), ER/PR status (n = 3273, p-value = 0.5), HER2 status (n = 3085, p-value = 0.5), or molecular subtype (n = 3273, p-value ≥ 0.6).

Overall, higher educational attainment was significantly associated with younger age at diagnosis, being employed, urban residence, and single marital status. Similarly, employment was significantly associated with urban residence and single marital status. Rural residence and single marital status was associated with younger age at diagnosis.

3.6. Time to Presentation

No significant association was observed between age at diagnosis and time to presentation (n = 3163, p-value > 0.1) (Table A1). However, patients aged over 60 had estimated odds of delaying presentation that ranged from 0.44 to 1.04 times those of patients under 40, with an upper confidence limit of 1.04, indicating a proximity to the null value of 1. Additionally, because the distribution of age differed significantly between patients with and without available data on presentation delay (Table A11), we conducted multiple imputation methods, as detailed in section 2.3.1. The results obtained after multiple imputation were consistent with those from the primary listwise deletion analysis (Table A14).

Patients with education above a diploma and time to presentation between 1-5 months showed a significant negative association (n = 3157, OR = 0.79, 95% CI: 0.67–0.94, p-value ≤ 0.01). Similar results were observed for patients who presented more than 5 months (n = 3157, OR = 0.69, 95% CI: 0.58–0.83, p-value ≤ 0.001), suggesting that patients with higher educational attainment had 21% lower odds of delayed presentation within1-5 months, and 31% lower odds of delayed presentation over 5 months compared to those with below diploma education (Table A2).

Additionally, employment status showed a significant association with time to presentation, where employed patients were significantly associated with a presentation time of 2–4 months (n = 3141, OR = 0.66, 95% CI: 0.53– 0.81, p-value ≤ 0.001) and greater than or equal to 5 months (n = 3141, OR = 0.69, 95% CI: 0.55–0.86, p-value ≤ 0.001), with the odds of shorter time to presentation being higher among employed compared to unemployed patients. However, for patients with a time to presentation greater than 5 months, this association lost statistical significance after adjusting for education level (n = 3136, p-value > 0.18), suggesting that the relationship between employment status and longer time to presentation may be confounded by educational attainment (Table A3). Furthermore, patients with rural place of residence were significantly associated with a time to presentation greater than or equal to 5 months (n = 3159, OR = 1.48, 95% CI: 1.12–1.96, p-value ≤ 0.01), indicating that patients residing in rural areas had 48% higher odds of presenting more than 5 months after symptom onset than those residing in urban areas. In contrast, there was no statistically significant difference in the odds of presenting within 1-5 months, compared with less than 1 month after symptom onset, according to place of residence (Table A4). No significant association was observed between marital status and time to presentation (n = 3161, p-value ≥ 0.8) (Table A5).

Given the large sample size of this study, the statistically significant differences in time to presentation observed across demographic groups may raise questions regarding their clinical relevance. However, the magni- tude of the observed differences should be interpreted within a population health context, where even modest differences in presentation delay may have important implications. The observed associations may reflect vari- ations in healthcare-seeking behavior, access to care, or structural and cultural factors related to presentation. Further research is warranted to better understand the mechanisms underlying these associations and to inform strategies aimed at promoting timely presentation and reducing disparities in presentation delay.

3.7. Clinicopathological Variables

The relationships between HR status, HER2 status, and histological grade are well established in the literature, and the findings of this study reflected these well-recognized patterns. ER positivity was strongly associated with PR positivity and HER2 negativity. Likewise, higher histological grade was positively associated with HR negativity and HER2 positivity.

In addition, the observed positive associations between invasive lobular carcinoma and ER positivity, PR positivity, and HER2 negativity are consistent with the known biological characteristics of invasive lobular carcinoma. Likewise, our findings also indicated that invasive lobular carcinoma was more positively associated with well and moderately differentiated tumors than with poorly differentiated tumors. This pattern is also in agreement with previous studies reporting that invasive lobular carcinomas are more likely to present as moderately differentiated tumors compared with invasive ductal carcinomas.^43,44 The significant differences seen in clinicopathological variables by ER/PR intensity appear to have no clinical relevance and are largely driven by the large sample sizes. The associations among hormone receptor status, HER2 status, histological grade, and tumor histologic types are presented in Tables A6, A7, A8, and A9.

Right-sided tumor laterality was initially associated with well differentiated grade (n = 3599, OR = 0.74, 95% CI: 0.58–0.93, p-value ≤ 0.05), however, this association lost statistical significance upon adjustment for ER/PR status (n = 2971, p-value > 0.2), suggesting that the relationship between laterality and grade may be confounded by hormone receptor status. The VIF for both ER and grade was 1.07, indicating low multicollinearity between these variables. Laterality showed no significant association with ER status (n = 3212, p-value > 0.3) (Table A6). No significant association between laterality and PR expression were seen (Table A7), consistent with the pattern observed for ER status and laterality.

Additionally, grade and number of tumors showed no statistically significant association (n = 2889, p-value > 0.8) (Table A9), likewise, no statistically significant association was observed between HER2 status and either number of tumors (n = 2412, p-value > 0.8) or laterality (n = 3035, p-value > 0.6) (Table A8). In contrast, patients presenting with multiple number of tumors was significantly associated with ER positivity, (n = 2549, OR =1.42, 95% = 1.01-1.99, p-value ≤ 0.001) (Table A6). Notably, the lower bound of the confidence interval lay very close to 1.00, yet the overall estimate remained statistically significant. The observed association between multiple tumor status and ER positivity persisted after adjusted with age at diagnosis. The association between PR status and the number of tumors was similar to that between ER status and the number of tumors, as shown in Table A7. This association was further investigated by multiple imputation, while multiple tumors remained significantly associated with PR positivity after imputation in crude model and in model adjusted by age at diagnosis (Table A13), the results should be interpreted with caution. Nonetheless, this finding may carry clinical relevance, the number of tumors is an important factor in surgical planning, influencing the choice between breast-conserving surgery and mastectomy, as well as decisions regarding systemic treatment. From a population perspective, this association may warrant closer attention to the assessment of tumor number in ER-positive patients during diagnostic phase.

4. Discussion

Considering age at diagnosis, the mean (SD) age was 50.68 (11.39) years, with a range of 16.9 to 90.4 years old. The median age [IQR] was 49.5 [42.3–58.4] years (Figure 2). The mean age in our study was higher than that reported in a neighboring country, Afghanistan, where the mean (SD) age at diagnosis was 42.9 (14.7) years, with ages ranging from 20 to 80 years. ⁴⁵ When compared with data from the United States, where the mean (SD) age of breast cancer patients was 60.91 (13.36) years (range: 18–90 years), the mean age of patients in our dataset was younger. Similarly, the proportion of younger patients in our cohort was higher than that reported in this study, in which approximately 6% of breast cancer patients were under 40 years of age, about 44% were between 40 and 60 years, and nearly 50% were over 60 years old. ²² In contrast, in our cohort, 17.22% of patients were under 40 years, 61.11% were between 40 and 60 years, and 21.65% were over 60 years. Our results are broadly consistent with previous studies conducted in Iran. Akbari et al. ²⁴ reported that the typical breast cancer patient was 40–50 years old. Haghpanah et al. found that the most prevalent age group was 40–55 years. ¹³ The predominance of breast cancer diagnoses among women under 60 is clinically concerning because it suggests earlier onset disease, which is often biologically more aggressive and carries longer term treatment-related morbidity.^46,47 Younger patients face distinct psychosocial and survivorship challenges including fertility preservation, career disruption, and caregiving responsibilities.^48,49

The median time from symptom awareness to first medical contact was 1 month, which suggested that half of patients with breast symptoms present within one month. However, the wide range (0.1–13 months) and positive skew indicated substantial heterogeneity. Notably, 25% of patients delayed more than 4 months, with some waiting over a year. This prolonged interval was significant, as diagnosis delays beyond 3 months have been associated with poorer prognosis in breast cancer. ⁵⁰ In the present study, 1052 of 3163 patients (33.26%) had a presentation delay of three months or more. This proportion was lower than that reported in Yogyakarta, Indonesia, where 43% of breast cancer patients experienced a delay exceeding three months. ⁵¹

Furthermore, the results showed that patients with lower educational attainment tended to delay presentation to a physician more often than educated patients. The persistence of this issue is particularly concerning given that, as early as 2003, a study conducted in Tehran, Iran, reported that 25% of breast cancer patients delayed presentation by more than three months and that lower educational status was positively associated with such delays. ⁵² Given the sociocultural and demographic similarities between Tehran and northeastern Iran, these findings are comparable to our results. Despite two decades of public health efforts, approximately one in three breast cancer patients, most of whom were less educated, still presented with delays exceeding three months. Likewise, patients living in rural areas more frequently experienced presentation delays longer than five months. This finding contrasted with a study conducted on patients in Victoria, Australia, which found that rural patients received treatment slightly faster than urban patients. ⁵³ It is worth noting that our dataset was collected during the COVID-19 pandemic, a period associated with major disruptions in healthcare access, delayed medical consultations, and reduced participation in cancer screening programs. Further studies are therefore needed to distinguish the specific impact of the pandemic from preexisting factors contributing to delayed breast cancer presentation.

Regarding demographic characteristics, the majority of patients in our study were from urban areas (3617/4027 (89.8%)), married (3852/4031 (95.6%)), and unemployed (3114/3970 (78.4%)). Akbari et al. noted that most breast cancer patients were married (94%) and had education diploma and above (68%). ²⁴ In contrast, our cohort showed a more balanced distribution between patients with a diploma or higher and those with lower educational attainment (2008/4014 (50.02%)). Education level, employment status, marital status, and place of residence can be used to assess socioeconomic status among breast cancer patients. However, the reported association of these factors with cancer characteristics is conflicting. While previous research has found that a higher education level was associated with an increased risk of developing breast cancer, ⁵⁴ our analysis showed that education was weakly associated with clinicopathological characteristics. Similar findings were observed for employment status, where any observed association with clinicopathological characteristics were no longer statistically significant after adjustment for ER/PR status, or time to presentation. Furthermore, urban–rural place of residence was not significantly associated with any clinicopathological variables. Additionally, patients residing in rural areas were more likely to be under 40 years old. The relationship between marital status and characteristics of breast cancer is also less well investigated in the literature. Single patients in our dataset tended to be younger, more likely to be employed, and more highly educated. We found that single patients were more likely to present with in situ histology, compared to invasive ductal histology. However, we were unable to adjust for age, education, or employment status due to violations of the log-linear model assumptions.

In the present dataset, tumors slightly more often affected the left breast 2054/3948 (52%) than the right 1879/3948 (47.6%) with the ratio of left to right side tumor 1.09, which is in line with study conducted by ²² who reported 50.6% for the left and 49% for the right breast. Similarly, Akbari et al. ²⁴ reported that 50.1% of breast cancers were left-sided, 49.0% were right-sided. No association between tumor laterality and any demographic or clinicopathological characteristics was seen, in line with with previous report by. ⁵⁵

Most patients in our study presented with a single tumor 2894/3089 (93.7%), while 195/3089 (6.3%) had mul tiple tumors. Our dataset classified tumors according to the number of lesions rather than explicitly distinguishing MF from MC disease. In a study conducted by Kanumuri et al., 82.3% of breast cancer cases were unifocal, whereas 11.3% were MF and 6.4% were MC. ⁵⁶ However, a direct comparison with our findings should be interpreted with caution due to differences in how tumor multiplicity was defined. Another study reported that the prevalence of MF and MC breast cancers varies widely, ranging from 1% to 60%, a range that underscores the sensitivity of these estimates to differences in definitions, imaging modalities, and pathological assessment meth- ods. ⁵⁷ Patients aged 40–60 years were less likely to have multiple tumors compared to patients under 40, which is consistent with previous studies reporting that younger patients were positively associated with multifocal and multicentric (MF/MC) breast cancers. ^{58 59} Nevertheless, tumor grade and HER2 status were not significantly associated with the number of tumors in our study, in contrast to, ⁵⁹ who reported a significantly lower HER2 positivity rate in multifocal/multicentric breast cancer compared with unifocal breast cancer but no significant difference in ER/PR expression between the two groups. Sensitivity analysis also confirmed these results (Table A13). No significant association was identified between demographic variables such as education, employment, marital and place of residence, and number of tumors overall.

Regarding tumor grade, among patients with available information, 2020/3639 (55.6%) had grade II tumors, 1301/3639 (35.8%) had grade III, and approximately 318/3639 (8.7%) had grade I. Similarly, Akbari et al. ²⁴ reported that 54.6% of tumors were grade II, 34.0% were grade III, and 12.0% were grade I. This distribution differs slightly from global statistics, which report approximately 43.4% of patients as grade II, 33.9% as Grade III, and 21.7% as grade I. ²² However, the order remains the same, with moderately differentiated tumors most frequent, followed by poorly differentiated, and then well differentiated tumors.

Akbari et al. ²⁴ found that 88% of tumors were invasive ductal carcinoma, 8% were invasive lobular, and nearly 5% were in situ, almost the same as our finding which 3541/4001 (88.5%) patients were diagnosed with invasive ductal carcinoma, 183/4001 (4.6%) invasive lobular, 104/4001 (2.6%) in situ. Pourriahi et al. ²⁵ also reported that 85.7% of tumors were invasive ductal carcinoma. The predominance of invasive ductal carcinoma in our dataset is consistent with global epidemiological patterns, highlighting its status as the most frequent histological type. ²² Mixed histology also consisted of 141/4001 (3.5%) in our data set, which was not reported in the comparative studies, a discrepancy that may suggest differences in detection practices used for classification.

Additionally, the lower proportion of in situ carcinomas in our study compared to some benchmark studies may similarly reflect our dataset’s limitation, as in situ carcinomas are frequently managed with surgery and systemic treatment alone without radiotherapy, and would therefore be underrepresented in this dataset.

Akbari et al. ²⁴ reported 70% ER positive, 67% PR positive, and 45% HER2 negative, while a smaller proportion of patients, 2043/3275 (62.4%) in our dataset were ER positive, and 1751/3275 (53.5%) were PR positive. Conversely, our cohort had a markedly higher rate of HER2-negative cases, 2330/3087 (75.5%), which may be because we classified equivocal HER2 status as either negative or positive according to FISH and CISH criteria, rather than reporting them as a separate category as Akbari et al. did.

A study conducted on Sudanese patients ⁶⁰ reported that triple-negative tumors accounted for 21%, Luminal A and B tumors represented 65%, and HER2-enriched tumors 14%. These findings were comparable to our results, which showed 2091/3275 (63.87%) Luminal A and B subtypes, followed by triple-negative tumors 788/3275 (24.06%) and HER2-enriched tumors 396/3275 (12.1%). Pourriahi et al. ²⁵ reported that 75.7% of cases were of the Luminal A and B molecular subtypes, 14% were triple-negative, and 10.3% were HER2-enriched. The proportion of triple-negative tumors in our cohort was higher than that reported by Pourriahi et al. ²⁵ The distribution of molecular subtypes in our study also differed from the findings reported by, ³⁹ who identified HER2-overexpressing tumors as the second most common subtype (10%–20%). The higher prevalence of triple- negative breast cancer in our study relative to these studies is clinically significant because triple-negative tumors are associated with more aggressive behavior and require complementary treatments, including radiotherapy in most cases.

Studies have shown that rather than focusing solely on cancer treatment after diagnosis, greater emphasis should be placed on prevention by addressing the underlying risk factors that contribute to disease develop- ment. ⁶¹ The findings of the present study highlighted the interplay between demographic, socioeconomic, and clinicopathological characteristics of breast cancer patients in northeastern Iran. By identifying these interrelationships, our analysis revealed patterns that could inform targeted preventive efforts to reduce the burden of breast cancer in this region.

4.1. Limitation

While our dataset offers rich and comprehensive information on breast cancer patients in northeast Iran, it has certain limitations.

• • In this study, the primary analyses were conducted using complete-case analysis, which may introduce bias and reduce statistical efficiency by excluding observations with missing data. Multiple imputation was also conducted as a sensitivity check for variables with high proportions of missing values (18–24%). Nonetheless, imputation results may be sensitive to model specification and the choice of variables included in the imputation model. Therefore, caution is warranted when interpreting findings for these variables.

• • Discrepancies in medical reports across different laboratories may complicate the accurate extraction of information. Additionally, variation in pathological scoring evaluation may introduce measurement error and bias the observed associations between variables such as ER/PR intensity and other variables. Furthermore, some variables relied on patient self-reports, may be affected by recall bias or misinterpretation.

• • The multivariable analyses primarily adjusted for age and hormone receptor status, and did not account for other clinically relevant confounders, including tumor size, lymph node involvement, treatment characteristics, and additional variables. As a result, the reported associations may be subject to residual confounding.

• • Given the large number of statistical comparisons performed, some statistically significant findings should be interpreted with caution, as they may represent false-positive results.

• • Patients referred to a radiotherapy oncology center may differ systematically from those managed exclusively in surgical or medical oncology settings with respect to disease stage, treatment pathways, and other clinical characteristics. Furthermore, underserved and underprivileged populations may be more likely to receive care through public healthcare centers and therefore may be underrepresented in the current cohort. Consequently, the study population may not fully represent the broader spectrum of breast cancer patients in the region, which may limit the generalizability of our findings.

• • As noted, most patients visited RROC to receive radiotherapy. Although patient status was tracked for up to one year after radiotherapy treatment completed, there is a possibility that some patients may have received recurrence treatment at other centers. Therefore, information on patient recurrence status and 5-year disease- free survival time is not fully available or transparent.

4.2. Future Research

The dataset and findings of the present study provide a foundation for future research. However, incorporating data from additional centers across the region would be valuable to improve representativeness and strengthen generalizability, particularly in light of the single-center referral bias noted above.

• • More studies are needed to determine underlying causes of observed associations. For instance, while patients residing in rural areas in our study tended to be younger, this observation does not imply that rural residence causes differences in age distribution of patients. The mechanisms underlying this pattern warrant further investigation. Incorporating variables such as tumor size, lymph node involvement, treatment characteristics, lactation history, and lifestyle factors into future studies may provide further insight.

• • Certain variables deserve further investigation within our dataset. Exploring these variables in more detail, possibly through additional modeling and method, could provide deeper insights into underlying patterns and relationships.

• • Although missing values are drawbacks of this study, they also provide an opportunity for future research in this area. Evaluating well-known imputation methods and other approaches for handling missing data using this dataset could be a valuable direction for future investigation.

• • Because part of this dataset was collected during the COVID-19 pandemic, future studies could specifically examine if the pandemic influenced the demographic and clinicopathological profile of patients presenting for treatment, including potential shifts in presentation timing across different demographic groups.

• • Considering patient survival times and applying methods specifically designed for survival data analysis, such as the widely used Cox regression model, would be an obvious next step. Exploring differences between groups at various follow-up periods and investigating various approaches to breast cancer survival in the presence of competing risks (CR), e.g., survived, local recurrence, or distant metastasis represent other directions for future research.

5. Conclusion

This study examined the demographic and clinicopathological characteristics of a cohort of female breast cancer patients treated at a radiotherapy and oncology center in northeastern Iran, addressing three research questions. Because complete-case analysis was used to handle missing data, sample sizes varied across individual analyses, and where relevant, age at diagnosis and ER/PR status were included as adjustment variables to clarify the observed relationships.

Regarding the first question, the initial cohort comprised 4034 female patients, among whom most were unemployed, urban-residing, and married, with educational attainment roughly evenly split between below diploma and diploma or above levels. The distributions of age, tumor grade, laterality, histology, and receptor status were broadly consistent with prior Iranian cohorts, while some patterns, including a higher proportion of triple negative tumors than reported elsewhere diverged from national comparisons.

With respect to the second question, socioeconomic factors showed limited associations with clinical outcomes, associations between education, employment status, place of residence, or marital status with tumor grade, receptor status, or molecular subtype were substantially attenuated after adjustment for age at diagnosis or ER/PR status . Age at diagnosis and its association with other variables followed the established pattern in the literature, where younger age was associated with multiple tumors, ER/PR-negative status, HER2-positive, poorly differentiated grade, Lobular subtype, and triple-negative subtype. This pattern remains even after adjustment for ER or PR status, but the 40–60 age group loses significance in most associations after adjustment.

For the third question, socioeconomic factors were more consistently linked to delayed presentation. Lower educational attainment, unemployment, and rural residence were each associated with longer time to presentation.

The associations identified here may in part reflect unmeasured demographic, biological, or environmental factors, and further research is needed to clarify the underlying mechanisms. Given the single- center design of this study, the generalizability of these findings to other populations should be considered with appropriate caution.

ORCID iD

Narges Motalebi https://orcid.org/0000-0001-6737-8890

Ethical Considerations

All assessments were conducted in accordance with the ethical guidelines approved by the Ethics Committee of Reza Radiotherapy and Oncology Center (RROC-1402-ETH-01).

Consent for Publication

All participants were informed, and their written consent was obtained.

Data Availability Statement

The raw data that support the findings of this study are available from the third author, H.K.H, upon reasonable request. The code used in this study is openly available on GitHub at: https://github.com/NargesMotalebi/Pre-Processing-RROC-Breast-cancer-Data.