Characteristics and outcome of breast cancer-related microangiopathic haemolytic anaemia: a multicentre study

The presence of schistocytes (> 0.5%) and either low haptoglobin or cytopenia (anaemia, thrombocytopenia or both) were mandatory for a diagnosis of MAHA, in agreement with current guidelines [15, 16]. Other eligibility criteria were patients with histologically proven breast cancer and MAHA diagnosed between 1995 and 2019 in participating centres. Ineligibility criteria were MAHA attributed to a cause other than breast cancer determined by treating physician and patients treated with either gemcitabine or bevacizumab at the time of or during the 6 months prior to MAHA diagnosis.

In July 2018, a call for participants was sent to 13 French cancer centres, outlining the study’s objectives, and listing the data that needed to be collected for the study. Participating centres were encouraged, whenever possible, to automatically screen their patient files by means of computerised searches using the following key words (in French): “microangiopathie(s) thrombotique(s); micro-angiopathie(s) thrombotique(s); micro angiopathie(s) thrombotique(s); schistocyte(s)”. Computerised screening of laboratory registries was also performed, searching for blood counts with elevated schistocytes. Senior medical oncologists manually reviewed all cases retrieved by computerised search to confirm the diagnosis of breast cancer-related MAHA. In centres in which computerised screening was deemed unfeasible, physician-based case declaration was also accepted. Case collection was closed in February 2019.

Data requested from participating centres are listed in Additional file 1, Supp Mat 1A. The PRONOPALL score, a validated prognostic score in oncology patients [17], was obtained from collected data. Because of the rarity of breast cancer-related MAHA, no data apart from those required for the diagnosis of MAHA were considered to be mandatory. The call for participants indicated that the study would investigate patient characteristics, response to treatment, outcomes and prognostic factors. In the absence of robust breast cancer-related MAHA data in the literature, no hypothesis could be formulated concerning the number of cases needed to achieve any of the study’s objectives; we did not hierarchise objectives into primary or secondary and this exploratory study did not have a predefined power.

The patients’ clinical characteristics are expressed as numbers and proportions; the Chi-squared test or Fisher’s exact test was used to compare categorical variables. Median follow-up was estimated using the Kaplan-Meier method.

Overall survival (OS) was determined from the date of MAHA diagnosis until the date of death or last follow-up. Survival curves were established by the Kaplan-Meier method. Biological and clinical factors were tested using a log-rank test in univariate analysis. The survival time variable was binarised into 2 categories: “death before or at 28 days” versus “death after 28 days”. The Hmisc package was used for imputation of missing data using the “aregImpute” function and to perform univariate and multivariate logistic regressions using “fit.mult.impute” with “lrm” as modelling function [18, 19]. Factors considered useful according to clinical considerations or with p value less than 0.2 in univariate analysis were included in a stepwise top-down procedure using the Akaike information criterion (AIC) and the likelihood ratio test as a criterion for variable selection. The accuracy of the final model was verified by controlling calibration and discrimation with the RMS package. The model with the lower value of Brier Score and the higher value of R² was selected as the final model to ensure the best discrimination. Calibration was assessed by visual examination of the calibration plot generated after bootstrap resampling. A prognostic score was constructed and weighted with β-coefficients estimation in the final model. The discriminatory capacity of the score, which represents the probability of dying within 28 days after the diagnosis of MAHA, was estimated by calculating the sensitivity and specificity of the score with their 95% confidence intervals (95% CI). A ROC curve was displayed and the area under the curve (AUC) was calculated using the ROCit package [20]. The 95% CI of the AUC was estimated by bootstrap.

All statistical analyses were performed with R software (version 3.6.2). This report was written in accordance with the REMARK guidelines.

Fifty-four cases of breast cancer-related MAHA from seven centres were included in this study. Forty-four cases were retrieved from four centres by means of computerised search, and ten cases were submitted by three centres based on the physician’s memory (Fig. 1). All patients were female. Median follow-up was 30.2 months (range 1.8–34).

Primary tumour characteristics, and clinical and laboratory features at MAHA diagnosis are shown in Tables 1 and 2, respectively. Invasive lobular adenocarcinoma or mixed adenocarcinoma with overt lobular component tumours was observed in N = 23 patients (44.2%). Few breast cancers were high-grade tumours (grade III, N = 13, 25.0%). Oestrogen receptor-positive/HER2-negative, HER2-positive, and triple-negative phenotypes accounted for N = 33 (68.7%), N = 7 (14.6%) and N = 8 (16.7%) breast cancers, respectively.

At breast cancer-related MAHA diagnosis, median age was 57 years (range 33–91) and all patients displayed breast cancer metastases. The median interval between diagnosis of breast cancer metastases and onset of breast cancer-related MAHA was 16.7 months (range 0–143.6); Kaplan-Meier Survival curves corresponding to time from the first cancer diagnosis and from the first metastasis until the development of MAHA are presented in Additional file 2, Supp Mat 2. None of the patients received mitomycin-C, gemcitabine or bevacizumab at the time or during the 6 months prior to MAHA for metastatic breast cancer. In N = 15 patients (28.0%), breast cancer-related MAHA was diagnosed either simultaneously or within 2 months of the diagnosis of metastatic breast cancer. Twenty-nine patients (53.7%) had three or more metastatic sites. Metastatic sites were mainly the bone (N = 44, 81.5%) and liver (N = 35, 64.8%), followed by the bone marrow (N = 11, 78.5%) and lung (N = 8, 14.8%). At MAHA diagnosis, 30 patients (55.6%) had a performance status (PS) ≤ 2; 9 patients (16.7%) presented clinical features of bleeding (ecchymosis, purpura, epistaxis, haematoma, haematuria and/or brain haemorrhage); 11 (20.8%) experienced dyspnoea, and 15 patients (28.3%) presented neurological symptoms (confusion, headache, dizziness, gait disorders, aphasia and/or somnolence), unrelated to presence of brain metastasis (not shown).

All patients had thrombocytopenia and anaemia. For 14 patients, elevated schistocytes (mandatory inclusion criterium) was reported in medical files, but exact counts were not available. Among 40 patients with available schistocyte counts, six (15.0%), 24 (25.0%) and 10 (25.0%) patients had schistocyte counts of 0.5 to 0.9%, 1.0 to 4.9% and 5.0% or higher, respectively. The clinical characteristics and outcome of the six patients with the lowest schistocyte counts were not significantly different from those of the other patients (not shown). Erythroblastemia and myelemia were commonly observed in 40 (85.1%) and 38 (90.5%) patients, respectively. Coagulation disorders (prothrombin time < 50%, platelets < 50G/L, fibrinogen < 1 g/L) were observed in six patients (11%), suggesting possible disseminated intravascular coagulation (DIC), according to the ISTH-DIC score (at least a score of 5) [21]. Impaired renal function, defined as glomerular filtration rate < 60 ml/min, was observed in 13 patients (34.2%). Thirty-six (69.2%) patients had elevated total bilirubin (≥ 1.24 mg/dL); Elevated bilirubin was not associated with presence of liver metastasis (Khi 2 test, p = 0.18). Other laboratory parameters at breast cancer-related MAHA diagnosis are shown in Table 2. Clinicopathological characteristics were not significantly different between cases retrieved by in silico screening and those reported by physicians (not shown).

Median overall survival (OS) was 4.0 weeks (95% CI [2.3;10.7]); Fig. 2a).Three-, 6- and 12-month survival rates were 32.6% (95% CI [22.1; 48.1]), 25.6% (95% CI [15.9; 41.2]) and 12.8% (95% CI [5.7; 28.4]), respectively. The 23 patients who received no anti-tumour therapy after MAHA diagnosis had a median OS of 10 days (range 0–203; Q1: 5.5; Q3:18). Twenty-one patients received one line of anti-tumour therapy and achieved a median OS of 47 days (range 0–353; Q1: 15; Q3: 98). Ten patients received two or more successive lines of anti-tumour therapy and achieved a median OS of 290 days (range 52–1035; Q1: 149; Q3: 524). Only yen patients were still alive after 6 months, and only 5 were still alive at or after 1 year. Noteworthy, outcomes are dramatically poor compared to general metastatic breast cancer population without TMA [22]. To identify patients that could be eligible for palliative care, we then assessed factors associated with overall survival less than 4 weeks. In univariate logistic regression analysis, PS 3/4 (odds ratio (OR) = 6.0, 95% CI [1.8; 19.8]), one or more prior lines of treatment (OR = 2.9 [1.0; 8.7]), elevated bilirubin (OR = 7.3 [7.8; 30.6]), haemoglobin < 8 g/dL (OR = 4.0 [1.3; 12.5]), prothrombin time < 50% (OR = 4.5 [0.9; 25.0]) and a short survival according to the PRONOPALL score (after implementation of missing data) (OR = 3.7 [0.8; 16.8]) were associated with a higher risk of death within 4 weeks of MAHA diagnosis (Table 2). Fourteen clinical characteristics among the 24 available were included in the multivariate analysis based on their p value less than 0.20 or for clinical rationale (Additional file 1, Supp Mat 1B). In multivariate analysis, PS 3/4 (OR = 7.0 [1.6;31.8]), elevated bilirubin (OR = 6.9 [1.1;42.6]), haemoglobin < 8 g/dL (OR = 3.7 [0.9; 16.7]) and prothrombin time < 50% (OR = 9.1 [1.2; 50.0]) remained significantly associated with a higher risk of death within 4 weeks of MAHA diagnosis (Table 2). The corresponding survival curves are shown in Fig. 2b–e. The PRONOPALL score being not an independent prognostic factor for early death, we combined the independent prognostic factors into a breast cancer MAHA survival score (Fig. 3a). Applied to our retrospective cohort, this score displayed a sensitivity for early death (< 4 weeks OS) of 0.86 (95% CI [0.67; 0.96]), a specificity of 0.73 (95% CI [0.52; 0.88]) and an area under the curve (AUC) of 0.90 (95% CI [0.83; 0.97], Fig. 3).