Therapeutic importance of timely immunophenotyping of breast cancer in a resource-constrained setting: a retrospective hospital-based cohort study

Background The Mexican Consensus of 2015 on diagnosis and treatment of breast cancer proposes the use of surrogate classification based on immunohistochemistry (IHC). In this study, we assessed the therapeutic value of this consensus recommendation, in the absence of gene expression data, as a tool for providing opportune and precise care. Methods and Findings In this retrospective hospital-based cohort study, we reviewed medical records of all suspected breast cancer patients in 2014 treated at the General Regional Hospital No. 1, of the Mexican Institute of Social Security (IMSS), located in Charo, Michoacan (western Mexico). Then, we followed the medical history of those patients with IHC testing until March 2017. The trajectory of patients was recorded to discover common patterns, and turnaround time for IHC testing. There were 402 suspected breast cancer patients in 2014, of which 30 had been tested for some IHC biomarkers (ER, PR, HER2). The surrogate subtyping allowed doctors to adjust (56.7 %) or confirm (43.3 %) the initial therapeutic regimen. Opportune IHC testing was found to be beneficial when it was available before or during the first rounds of chemotherapy. Conclusions The use of opportune immunohistochemistry provides improved outcomes for breast cancer patients. This result has the potential to become a roadmap for the implementation of precision medicine in Mexico. Abbreviations IMSS, Mexican Institute of Social Security; HGR1, General Regional Hospital No. 1 in Charo, Michoacan, Mexico.


Introduction
been previously recognized to have prognostic value in the management and treatment of breast 48 cancer patients [1]. During the last decade, there have been several commercially accepted multi-49 gene prognostic tests. Genomic prognostic tests have been associated with decreased adjuvant 50 chemotherapy use among patients, and better outcomes with cost-effective treatments [2]. 51 However, the availability and expense of these genomic tests remain largely prohibitive for 52 constrain-resourced hospital settings, regardless of country income. For example, in the United 53 States, the use of the 21-gene recurrence score Oncotype DX has increased significantly over the 54 last decade due to insurance coverage, but it is estimated that only a quarter of eligible patients 55 have been tested [3]. 56 In resource-constrained settings, the use of surrogate immunohistochemistry (IHC)-based 57 breast cancer subtypes presents as an alternative to molecular subtyping when there is no access 58 to gene expression data or high-throughput sequencing data. Despite the lowering costs of 59 genomic tests, these technologies are only available to researchers in select locations in the 60 world, and they are inaccessible to most patients worldwide. Alternatively, a panel of three IHC 61 markers-estrogen receptor (ER), progesterone receptor (PR), and human epidermal growth 62 factor receptor 2 (HER2)-demonstrated its predictive value for chemotherapy response in 63 breast cancer [4]. There are other IHC markers that could be included for the prediction of 64 treatment response and prognosis; however, further studies are needed to select an optimized 65 surrogate panel [5]. 66 The 2015 Mexican Consensus on diagnosis and treatment of breast cancer [6] is the main 67 working document for oncologists in the country. It recommends routine use of a 3-marker panel 68 (ER, PR, HER2) IHC approximation to identify four intrinsic breast cancer phenotypes (luminal 69 A, luminal B, HER-2, and triple negative). In the Mexican population, the estimated frequency 70 of ER/PR positive is 60%, HER2 positive 20.4%, and triple negative is 23.1% [7,8]. There is a 71 troubling onset of disease among younger breast cancer patients in Mexico (aged <40), with a 72 high prevalence of triple-negative breast cancers [9]. Furthermore, there is an increased mortality 73 trend associated with breast cancer in the country [10]. 74 The Mexican clinical practice guidelines for screening, management, and treatment of 75 breast cancer [11][12][13] also recommends the use of IHC testing as part of the histopathology 76 studies. Despite these recommendations, the widespread use of genetic testing in Mexico remains 77 elusive due to constrained resources. However, IHC testing is widely available through hospital-78 owned pathology laboratories, as well as through outsourced commercial laboratories across the 79

country. 80
In this study, we aimed to assess the timeliness of surrogate IHC biomarker testing in the 81 context of a resource-constrained hospital setting. Evaluating the therapeutic value for patients is 82 of critical importance, since the demand for more IHC testing will create new logistical 83 challenges in the healthcare system. The impact of acceptable or critical time delays is ultimately 84 not well defined, and it would be of great importance for hospital administrators to have a 85 reference of the value of timely IHC testing. As precision medicine continues to provide better 86 prognostic biomarkers, translating these benefits into clinical practice increasingly requires a 87 time-efficient approach from the entire healthcare system. 88 breast cancer who require radiation therapy are provided the service through an outsourced 114 private service in the same city. The personnel include gynecologists, medical oncologist, 115 surgical oncologists with significant breast cancer training, adequate numbers of nursing and 116 pharmacy staff, surgeons with significant training, and cancer pathologists. A self-assessment of 117 the services and facilities provided by these hospitals is S1 Text, which is based upon a modified 118 version of the oncological service self-assessment levels for low-and middle-income countries 119 developed by The International Society of Pediatric Oncology (SIOP) committee on Pediatric 120

Patients 123
The study population for this research was all cumulative breast cancer patients seen in 124 2014 by medical staff at the IMSS's General Regional Hospital No 1 in Charo, Michoacan. The 125 institution's breast cancer census was used to select patients that were seen in 2014 at HGR1. To 126 avoid selection bias in our study, we did not restrict the selection of patients to members of the 127 female sex or any age group. The inclusion criteria can be seen in Figure 1. A patient with suspected breast cancer is referred from a primary-care facility to HGR1 132 after being examined by his/her family physicians. If the physician suspects the existence of a 133 breast lesion, the patient undergoes an imaging assessment using the breast imaging-reporting 134 and data system (BIRADS) score. Upon arriving at HGR1, the patient was either seen by the 135 breast cancer clinic (BIRADS 0 and 3) for further imaging studies and/or fine-needle biopsy 136 assessments, or directly referred to the medical oncology service (BIRADS 4, 5 and 6), at which 137 point they would undergo surgical biopsies for pathology analysis. The gold-standard diagnosis 138 of breast cancer would be provided by a board-certified pathologist, who would assign one of the 139 C50 codes (malignant neoplasms of breast) from the international code of diseases version 10 140 We reviewed the medical records (both electronic and paper) of all breast cancer patients 142 at HGR1, and selected those whose biopsies had also been tested with IHC to detect ER, PR, 143 HER2 antibodies; then, we followed the medical histories of those patients, ending with March 144 2017, making note of multiple medical visits to the breast cancer clinic, medical oncology 145 service, and/or surgical oncology service. From their medical records, we extracted information 146 about IHC testing (antibody ordered, date of ordering, date of results being obtained), 147 chemotherapy and hormonal drugs administered, and radiation sessions and surgical procedures 148 undertaken. 149 150

Study design 151
This was a retrospective hospital-based cohort study, using medical records collected System (UMLS). When disagreement occurred, both authors discussed the translation to assign a 166 UMLS code to the medical records in Spanish. Finally, similar events were grouped together in 167 fewer categories. The UMLS codes, UMLS descriptions, Spanish description, and groupings can 168 be seen in Table 1. 169

172
Missing data. Missing data was assumed to be non-missing at random (NMAR), and the 173 events from Table 1 were considered missing for a specific reason. For example, if the code 174 C2186775 (request lab result from MRI) was missing from a patient's medical record, we left 175 that value unassigned, since it was assumed that an MRI was not requested for that patient. There 176 is a small chance that this assumption might be violated (and introduce some bias) if the 177 physician did request the exam/procedure but forgot to write it down in the medical record. 178 However, we did not impute any missing data, since that might be a larger source of bias for this 179 EHR-based data. 180 Outliers. We assessed the time-to-event distribution for each event in Table 1, and 181 reported the mean time using boxplots with whiskers. The outliers were identified using the 1.5 182 interquartile rule, but were not removed from the downstream analysis since undefined time-to-183 event standards exist for these events. 184 Clustering. A heatmap with all the patients and events was generated. We used 185 hierarchical clustering to create dendograms. The distance matrix was generated with the time-

198
The epidemiological information obtained pertaining to this one IMSS hospital can be 199 seen in Table 2

Clustering analysis 221
A hierarchical clustering and heatmap of time-to-events is shown in Figure 2. The time in 222 which each event occurred for each patient is measured using information found in their medical 223 record (in days, positive). To be able to create a dissimilarity matrix to calculate distance 224 between patients, all missing values were assigned a negative value of -1000. It is important to 225 note that this figure does not provide a context for the ordering in which events occurred, but it 226 can still help to provide information about patients with similar trajectories. Three main groups 227 of patients (or clusters) were identified in this graph. 228  The time-to-event for all events is shown in Figure 4. It is measured as the number of 252 days that it took for an event to occur, from the initial visit to the hospital stay (day 0). From this 253  The turnaround time for immunohistochemistry (IHC) testing can be seen in Figure 5. 263 Only 20 patients had information in their medical records relating to the timing of IHC testing. 264 Although this service is referred to in the IMSS medical records, it is subcontracted to a private 265 laboratory. There was large variation surrounding two key variables: a) the time needed to 266 request IHC testing relative to the first day of follow-up treatment at the hospital, and b) the

Association rules 286
Association rules were mined from the events of this cohort using the Apriori algorithm. 287 The five top-ranking association rules are shown in Figure 6. These rules have a support of 288 63.3% of cases, with a confidence of 90.5% and a rule performance (lift, or precision) of 1.3. For 289 rules with lift larger than 1, it can be assumed that the right-hand side of the rule is a 290 consequence of the left-hand side. In our mined rules, we can observe that, in general, IHC 291 testing, particularly ER/PR testing, results in adjusting treatment plans.

Key findings 298
Breast cancer in Mexico (and worldwide) continues to be a public health concern, and 299 increasing research is being done on the characterization of better genomic variants. Although 300 our study investigated a small number of patients, the prevalence and epidemiological 301 characteristics among our cohort are similar to what has previously been estimated for the 302 country (i.e. in regards to age group and staging) [10]. Our study further investigates the 303 timeliness of IHC testing, so our conclusions can then be used to inform hospital administrators 304 and public health officials around the world about the importance of timely IHC testing. 305 On the therapeutic value. The therapeutic importance of immunophenotyping (the use of 306 IHC testing for subtyping of cancer patients) needs to be carefully considered and addressed in 307 treatment plans by oncologists. IMSS protocol states that estrogen positive (ER+) and/or 308 progesterone positive (PR+) patients could benefit from hormone replacement therapy to prevent 309 recurrence. For the patient, that usually means that the traditional FEC chemotherapy regimen 310 composed by 5-fluorouracil, epirubicin, and cyclophosphamide, would be adjusted to include 311 some hormone replacement therapy, either aromatase inhibitors to block estrogen production 312 (e.g. anastrozole, letrozole, or exemestane), or some drug that could interfere with the ability of 313 estrogene to stimulate the growth of breast cells (e.g. tamoxifene or toremifene). 314 In contrast, patients with HER2 positive (HER2+), which tumors tend to grow and spread 315 more aggressively, could benefit from a targeted therapy that could prescribe drugs to block 316 HER2 (e.g. trastuzumab or pertuzumab). However, the use of these drugs can have different side 317 effects than traditional chemotherapy. The use of in situ hybridization (ISH), instead of IHC, can 318 be used to determine HER2 status with an overall concordance with IHC, and it may be more 319 beneficial to use both [18]. There is a need to balance sensitivity of the study with hospital 320 workload and expenses. Furthermore, HER2 status has been successfully incorporated into 321 medical practice to guide treatment decisions for breast cancer patients [19]. In fact, the 322 On the need for accurate electronic health records (EHR). Missing information was 363 common across the paper and electronic records in our study. Clinicians and administrative staff 364 at the IMSS hospital are still getting used to the novel implementation of the EHR, which had an 365 impact on our ability to better characterize this cohort. In the absence of a cancer registry, our 366 best estimate of the disease is the institutional census. As the hospitals in Mexico, and elsewhere, 367 continue to become more and more electronic, there is a need to develop better software tools to 368 analyze the information obtained, including medical natural language processing and machine 369 learning applications. 370 On the visual representation. The medical trajectories shown in Figure 1 are useful in 371 quickly providing a patient history overview of hospital care. With more work on the user 372 interface, we envision this tool could eventually represent a valuable visual aid in which a patient 373 and their companions might be able to better communicate with their clinicians about the 374 management of their disease (it could probably also appear in a printed version). Currently, some 375 of the challenges of this tool include the missing identification of events that are overlapping, 376 and the impossibility of further elaborating on the details of each event. For a hospital 377 administrator, Figure 1 can provide a quick overview of the patients seen in their hospital, which 378 could be used as a decision-making tool, with the proper validation. This visual aid creation tool 379 can be obtained from the repository at https://github.com/arturolp/patientTrajectory.git 380 381

Healthcare system implications 382
Adverse events in the trajectory of an oncological patient, which might include 383 hospitalization related to cancer, recurrence of tumor, or metastasis, are extremely costly for the 384 healthcare system. The use of IHC testing was shown in our study to help with the selection of 385 precise treatment for patients (either by adjusting the treatment or confirming it). The association 386 rules found in our study confirm that ER and PR testing are associated with the adjustment of 387 treatment. However, the time in which IHC testing is performed is of critical importance if we 388 want to influence improved prognosis. 389 IHC testing is being carried out as an external service at the IMSS hospital in our study. 390 As the Mexican healthcare system continues to transition from reactive to preventative care, the 391 need for more IHC testing in breast cancer and other diseases will certainly allow for the further 392 development of its own testing facilities, therefore allowing for some economy of scale. 393 We have shown that the use of surrogate immunohistochemistry, as described by the 394 Mexican Consensus and previous literature worldwide, can have a beneficial therapeutic effect 395 on breast cancer patients. The aims of any healthcare system should be the identification of 396 earlier events that can have an impact on downstream events in the trajectory of an oncological 397 patient's treatment. In resource-constrained settings, it is important not only to consider surrogate 398 alternatives to more costly diagnostics (e.g. genomic testing), but also to incorporate the 399 regulatory and logistical aspects of implementing these surrogate IHC tests. IMSS must face the Research reported in this publication was supported by grant 8316sc-7 from the joint 432

Center of Excellence in Regulatory Science and Innovation (CERSI) at the University of 433
California, San Francisco and Stanford University. The funder had no role in study design, data 434 collection and analysis, decision to publish, or preparation of the manuscript. 435

Competing interests 437
RAR declares to be IMSS's state delegate in Michoacan, overseeing the hospital 438 mentioned in this study. The remaining authors declare no conflicts of interest. 439