Are There Socio-Demographic Inequalities in the Utilisation of Tumour and ctDNA Somatic Mutation Testing in Solid Tumours? A Systematic Review.

1
Introduction
It is recognised that inequalities in access to diagnostic and therapeutic modalities due to socio‐demographic status affect patients with cancer [1]. Patients from deprived backgrounds are less likely to have definitive surgical management of malignancy and adjuvant chemotherapy or radiotherapy; less likely to receive biological or precision therapies or immunotherapy; more likely to face substantial financial toxicity in attending appointments; and ultimately more likely to die of their disease [2, 3, 4, 5]. Similar patterns in equality of access to existing treatments are well characterised for older patients and patients from different ethnic backgrounds, such as disparities between women's outcomes in breast cancer between ethnic groups in New Zealand [6, 7, 8].
The previous decade has demonstrated rapid expansion of availability of commercially available next generation sequencing (NGS) based comprehensive genomic profiling technology for somatic mutations in tumours [9]. A timeline of major transatlantic milestones within this field is shown in Figure 1. Somatic mutation testing has been developed across both solid tumour testing and cell free or circulating tumour DNA (ctDNA). The first such ctDNA panel with Food and Drug Administration (FDA) approval was the Guardant360 Companion Diagnostic (CDx) assay in August 2020, followed within weeks by Foundation One Liquid CDx—representing how recently commercial availability of this testing has been realised [10, 11]. NGS testing allows for targeting of cancer therapeutics and generates information for prognostication and enrolment into clinical trials. Testing, therefore, advances our ability to realise the potential of ‘precision oncology’ from within the clinic. Some forms of panel sequencing tests have been available for longer—since 2004 in the case of the 21‐gene panel recurrence test score in breast cancer, also known as Oncotype DX. This is a predictive genetic test which does not utilise NGS but reverse transcription polymerase chain reaction (RT‐PCR) of 21 recognised genes involved in breast cancer to stratify patients into those who are at highest risk of recurrence and are therefore most likely to benefit from adjuvant chemotherapy. In 2013, Oncotype DX additionally launched a 17 gene panel for use in prostate cancer [12]. Our understanding of the utilisation of Oncotype DX panels, as such data reaches maturity, is only contemporarily being investigated [13, 14].
Socio‐demographic disparities in the access to and utilisation of broad panel somatic tumour mutation testing remain largely unexplored. Novel modalities are associated with high purchase costs, which may be presumed to be barriers to reaching the most disadvantaged and marginalised patients with cancer, especially in non‐publicly funded healthcare settings, although many varied healthcare systems support and reimburse costs for such testing [15]. However, in more highly tailoring initial therapeutics, utilising molecular data, this may provide a pathway for reducing current treatment inequalities [16]. Previously, Norris et al. explored socio‐economic inequalities in utilisation of predictive biomarker tests and biological and precision therapies for cancer and found there to be socio‐economic inequalities in predictive biomarker tests and biological and precision therapy utilisation [17]; that review included studies published up to 2019, although many included patients diagnosed years previously.
This current systematic review collates new literature, specific to tumour and ctDNA somatic mutation testing of solid tumours in relation to broader socio‐demographic factors since 2018, as an update to our understanding of the rapidly evolving literature in this field.

2
Methods
This review was registered with PROSPERO, the international database of prospectively registered systematic reviews (CRD42023487689). It is reported in accordance with the Preferred Reporting Items for Systematic Reviews and Meta‐Analyses (PRISMA) 2020 statement (Table S1).
2.1
Search Strategy and Study Selection
Seven databases (MEDLINE, EMBASE, Scopus, CINAHL, Web of Science, PubMed and PsycINFO) were searched. Articles published between January 2018 and March 2025 were included. This time period was selected to reflect the period during which multiple comprehensive genomic sequencing panels received approvals and reimbursement worldwide.
Both tumour sample and ctDNA somatic mutation testing were reviewed across all solid tumours. Germline testing and testing done in non‐malignant tumours were excluded. Studies exploring single gene tests in isolation were not included as these represented an older technology licensed significantly prior to 2018. Search terms covering socio‐demographic status, disparities in utilisation of, and comprehensive genomic sequencing panels were developed. Full search strategies are included in Supplementary Methods (Table S2).
The inclusion criteria for full‐text studies written in English were determined by a defined PICOS question as follows:
2.1.1
Population
Solid malignant tumour diagnosis in a patient of any age or sex.

2.1.2
Intervention
Utilisation of a comprehensive genomic sequencing panel for somatic mutation testing during malignant tumour management.

2.1.3
Comparison
It was a requirement that a comparator was reported—where reported, this included—a clinical alternative or no comprehensive genomic sequencing panel being used.

2.1.4
Outcome
Utilisation data reported by a socio‐demographic measure (e.g., race/ethnicity, median household income, age, education level, sex/gender).

2.1.5
Setting
Retrospective or prospective observational study (including randomised controlled trials analysed as observational cohorts). A decision tree to determine eligibility of full texts is included in Figure S3.
Title and abstract screening was completed by one author (S.R.), with 10%+ confirmed between four authors (R.N., A.G., L.S. and A.B.). Disagreements were discussed and agreed with a senior author (A.G.). All full text studies were reviewed by one author (S.R.) and independently reviewed by at least one other author (R.N., A.G., A.B. or M.J.L.). Disagreements were discussed and if required resolved with the involvement of a third author (A.G.). Agreement between reviewers was excellent (κ = 0.96). Forward and backward citation searching was conducted for eligible studies as well as citation searching of previous reviews on related topics.

2.2
Data Extraction and Quality Assessment
Data was extracted by one author (S.R.) and confirmed by a second author (A.B. or M.J.L.). Disagreements were resolved through discussion with a third author (A.G.). Where data was unclear, the study authors were contacted by email. Studies were included only if responses met the inclusion criteria. Duplicate usage of the same database for overlapping studies was noted and considered in discussion.
Data extracted included: author(s); publication year; country; data source; number in study population; malignant solid tumour diagnosis; patient age(s); socio‐demographic measure(s) and unit (including measures of socio‐economic group); comparator(s) were included; and measures of association for not having testing completed using a comprehensive genomic sequencing panel by socio‐demographic status (e.g., OR, 95% CIs, p‐values, etc.). If adjusted and unadjusted analyses were reported, the adjusted results were extracted; if more than one multivariable analysis was conducted, information was extracted from the most comprehensive adjusted model.
All studies deemed eligible were quality appraised and scored using a modified version of the ISPOR checklist for retrospective database studies (Table S4) (modifications as per Norris et al.). Appraisal was completed by one reviewer (S.R.) and discussed with the review team (A.B., M.J.L., R.N., L.S. and A.G.).

2.3
Synthesis of Evidence
Data was synthesised using a summary of findings table. Where not reported (and where possible), percentages of utilisation of comprehensive genomic sequencing panels by socio‐demographic metric(s) were calculated from data reported in the included studies. Studies were noted to be heterogeneous in terms of outcome analyses, socio‐demographic comparisons made, whether ORs (crude or adjusted) were reported, and the variables that any adjusted ORs were controlled for. Unadjusted ORs were computed by the review authors when not reported to enable inclusion of as many studies as possible in the synthesis in a consistent way. The narrative synthesis divided by Oncotype DX testing and NGS testing (both tumour sample and ctDNA analysis) was completed for included studies. This division between Oncotype DX and NGS testing was completed due to recognition of variation in usage of testing modalities and differences in timeframe in which they were introduced to cancer care. Meta‐analyses of results were not conducted due to the small number of studies identified, and heterogeneity in socio‐demographic metrics and tests. Duplication of data sources further reduced those eligible for statistical combination.

3
Results
3.1
Search Results
The search identified 33,584 citations. There were 5835 duplicates removed and title and abstract screening of 27,749 records was completed to assess for eligibility. Of these, 60 studies were retrieved for full‐text review. Eight additional studies were identified by forward and backwards citation searching and citation searching of previous related reviews. Overall, 24 studies identified met the inclusion criteria, had appropriate comparators, and were included in the review (Figure 2, Table 1).

3.2
Study Characteristics
The included 24 studies covered two main testing modalities: Oncotype DX (n = 11 studies) [14, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36] and NGS sequencing technology (n = 13) [37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49]. Foundation One panels represented the most common NGS sequencing modalities named by included studies (n = 2) [41, 42]. Included NGS studies detailed six main primary tumour sites (non‐small cell lung cancer (NSCLC), gastrointestinal, gynaecological malignancies, pancreatic, prostate and breast cancer). Most studies were from the USA (n = 23) [14, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 43, 44, 45, 46, 47, 48, 49] with the remaining study from Israel [42]. Three studies on NGS reported data from the Flatiron Database in three tumour types—urological, colorectal and NSCLC, respectively [39, 45, 47]. One study from the Flatiron Database reported breast cancer in addition to colorectal and NSCLC data as one combined cohort [37]. Two studies on NGS utilisation reported data on NSCLC from the National Cancer Institutes' Surveillance, Epidemiology and End Results (SEER) population based Cancer Registry Programme [43, 48]. Three Breast Cancer Oncotype DX studies reported data from the National Cancer Database [14, 30, 34]. These reported data across differing stages and types of breast cancer with socio‐demographic factors defined by different subcategories (e.g., inclusion of Location: metro/urban/rural [14]; or education status [34]). Two Oncotype DX studies reported data from The Carolina Breast Cancer Study Cohort [28, 35].
Socio‐demographic measures of test utilisation were reported across 12 measures: race and/or ethnicity (n = 23) [14, 27, 28, 29, 30, 31, 32, 33, 34, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49] age (n = 16) [14, 27, 29, 30, 31, 32, 33, 34, 35, 36, 41, 42, 43, 45, 47, 48]; insurance status (n = 15) [14, 27, 29, 30, 31, 32, 33, 34, 35, 39, 41, 43, 45, 46, 47]; socio‐economic status (n = 13) [14, 27, 31, 33, 34, 35, 39, 43, 44, 46, 47, 48, 49]; sex (n = 8) [29, 43, 44, 45, 46, 47, 48, 49]; location (urban/rural, etc.) (n = 7) [14, 36, 43, 44, 46, 48, 49]; marital status (n = 5) [31, 33, 35, 43, 48]; education (n = 3) [34, 35, 46]; and hospital status (e.g., teaching hospital) (n = 3) [36, 41, 43]. Three socio‐demographic measures were only reported in one study each, these included—smoking status [47]; distance from large centre [29]; and disability status [43]. Four studies [28, 37, 38, 40] reported socio‐demographic information by one measure, race and/or ethnicity, only; the others all reported multiple measures. For one study [33] utilisation was available as a percentage only. One study reported hazard ratios (HRs), confidence intervals (CIs) and p‐values only [39]. The majority of studies (n = 23) featured no testing as a comparator, with one study comparing utilisation by size of panel (large and medium) against single gene testing [49]. No paediatric specific studies were retrieved.

3.3
Quality Appraisal
The 24 included studies quality scores using a modified ISPOR checklist ranged from 6.5 to 10 out of a possible 10 (mean = 8.3, median 8) (Table S5). The study populations, socio‐demographic factors, and numerical reporting were often described well in the included studies. Methods for adjusting for confounders and statistical analysis for associations between exposure (socio‐demographic metric) and outcomes (test utilisation) scored lower.

3.4
Socio‐Demographic Factors and Utilisation of Oncotype DX
The data reported in studies on Oncotype DX testing in both breast and prostate cancer was collected from 2004 to 2018. Increasing age was significantly associated with decreased likelihood of utilisation of Oncotype DX across seven of ten studies which reported age [27, 29, 30, 32, 34, 35, 36]. Race and ethnicity were reported across a diverse range of classifications. In seven of ten studies reporting on race and/or ethnicity—white or non‐Hispanic white was associated with an increased likelihood of receipt of testing [28, 30, 31, 32, 33, 34, 36]. Likewise, privately insured patients or fee‐based care (as opposed to Medicaid, Medicare, uninsured or non‐fee‐based care) were more likely to be in receipt of testing; this was reported in seven of eight studies which reported on insurance status [14, 30, 31, 32, 33, 34, 35].
Socio‐economic status (median income and education level) was reported in six studies; five demonstrated that increasing socio‐economic status was related to higher receipt of testing [14, 27, 31, 33, 35]. However, one study reported increased utilisation with lower socio‐economic status (based on median income) [34] as part of an adjusted OR model. Three studies reported on marital status, with two of them finding being single/unmarried or divorced/separated associated with a higher likelihood of Oncotype DX testing when compared to married patients [30, 31]. In one such study, widowed patients had a significantly reduced likelihood of testing (OR 0.79, 95% CI 0.69–0.89) [33]. As the breast cancer Oncotype DX testing modality is female predominant, only one study reported on utilisation by sex, in US veterans. This found low Oncotype DX utilisation in male compared to female breast cancer; however, the number of cases in males was small and the association was not statistically significant (n = 72/267 and 10/61 respectively, OR 0.53, 95% CI 0.26–1.10, p = 0.09) [29].

3.5
Socio‐Demographic Factors and Utilisation of NGS Testing
Thirteen included studies presented data on NGS testing across six main tumour types; pancreatic (n = 1) [40], prostate (n = 2) [39, 46], gastrointestinal (n = 4) [37, 44, 45, 46], gynaecological (n = 2) [41, 42], NSCLC (n = 8) [37, 38, 43, 44, 46, 47, 48, 49] and breast (n = 3) [37, 44, 46]. Data presented was collected from 2002 to 2023. All thirteen studies included reported race and/or ethnicity and nine studies showed an association with non‐white races and/or ethnicities and a decreased likelihood of receipt of NGS [37, 38, 39, 40, 41, 43, 44, 46, 47]. One study [42] reported ORs based on Ashkenazi Jewish origin (no vs. yes) and demonstrated a decreased likelihood of testing in patients not of Ashkenazi Jewish origin in Israel (OR 0.45, 95% CI 0.29–0.70, p = 0.0003). Three studies, in US populations, identified increased ORs for utilisation of testing in Asian populations when compared to white populations (OR: 1.54 [1.23–1.93] [43]; OR: 2.31 [1.40–3.81] [46]; OR: 1.15 [0.81–1.65] [47]). These studies all reported on populations exclusively consisting of or containing patients with NSCLC. Age was reported in five studies [41, 42, 43, 45, 47] and in two studies older age was associated with decreased utilisation [45, 47]. Two studies did not report ORs for age [41, 42]. One study [43] demonstrated inconsistency across age ranges reported, from 66 to 99 years old, with increased utilisation between ages 81–85 years old when compared to 66–70 years old (OR: 1.15 [0.94–1.42]).
Insurance status was reported in six studies [39, 41, 43, 45, 46, 47]; private insurance (as opposed to Medicaid, Medicare or uninsured) was associated with increased receipt of testing in all six studies. Socio‐economic status was reported in seven studies and lower socio‐economic status was associated with decreased utilisation in six of these [39, 43, 44, 47, 48, 49]. Sex was reported in seven studies [43, 44, 45, 46, 47, 48, 49]. Female sex was associated with increased utilisation in four studies on patients with NSCLC or containing patients with NSCLC [43, 46, 47, 48]. In one NSCLC study [47] a history of smoking was associated with reduced likelihood of testing when compared to never smokers (OR 0.63 [0.54–0.75], p < 0.001). Disability status was reported by one study. This found significantly decreased test utilisation in patients with advanced lung cancer and a poor disability status (OR: 0.61 [0.48–0.79], p < 0.001) [43].

4
Discussion
This systematic review represents the first to assess utilisation of novel tumour and ctDNA somatic mutation testing in solid tumour by socio‐demographic factors since 2018. The review serves as a timely update as significant advances in access to testing have been made in recent years. Across multiple tumour types, testing modalities and socio‐demographic characteristics, statistically significant disparities were identified. These were most comprehensively characterised by decreased test utilisation in: non‐white races and ethnicities; older age; those with non‐private insurance statuses; and less socio‐economically advantaged groups. These findings support the previous literature which has highlighted inequities across the cancer treatment pathway [1, 2, 3, 4, 6] as well as older reviews exploring disparities in access to novel biomarker guided therapy [5]; access to treatment; and survival [51].
Studies on the Oncotype DX testing panels in breast and prostate cancer represented slightly less than half of those identified. As a more established testing modality at time of publication [11], it was of interest to assess if disparities in access were present following over a decade of usage and familiarisation. Literature prior to inclusion criteria dates for this review had consistently identified disparities in Oncotype DX utilisation by race [52, 53, 54]; increasing age [55]; insurance status and reimbursement of test cost [56]. It might have been expected that, as both clinicians and patients became more familiar with the test, and it became more widely used, inequalities might disappear over time. In fact, the contemporary literature identified by the present review found broadly the same patterns, suggesting a lack of significant shift in the identified inequalities in the intervening time period. Indeed, inequities of access by race; socioeconomic group and location have been highlighted as recently as 2024 [35]. These findings are concerning, as low uptake of Oncotype DX testing in breast cancer has implications for chemotherapy receipt and guideline adherence for optimising patient care. One study was identified exploring disparities in utilisation of the Oncotype DX prostate cancer test [31]. As a more recent and less utilised testing modality, this demonstrated similar patterns of disparities in decreased utilisation with increased age; non‐white races and/or ethnicities; lower socioeconomic status; and non‐private insurance.
NGS testing panels were grouped separately for review. Similarly to the Oncotype DX studies, these indicated generally decreased utilisation in non‐white races and/or ethnicities; non‐private insurance status; lower socio‐economic groups; and potentially older age. When included, Asian or Asian/other ethnic groups demonstrated increased utilisation compared to white population comparators in NSCLC studies [43, 46, 47]. This may represent clinician bias in favouring genomic testing where certain socio‐demographic factors have shown evidence of increased likelihood of actionable mutations. Patients of Asian descent with lung cancer have been demonstrated to have a higher burden of EGFR and ALK actionable mutations [57]. Likewise, female gender was associated with increased utilisation in four studies [43, 46, 47, 48]. These studies were all on patients with NSCLC or contained patients with NSCLC. Certain targetable mutations, particularly those in EGFR, are recognised to be more prevalent in female populations [58]. Therefore, this may again demonstrate an element of clinical bias where NGS testing is selective.
In the USA, Medicare national coverage determination for reimbursement for NGS testing commenced in 2018. As such, insufficient time has passed for large numbers of patients benefiting from such to have been included in studies in this review. Reimbursement and the large amount of NGS products now available on markets may aid in utilisation among patients without private insurance in the United States (US) population. In the present review, the cohorts included feature data often from the transition time prior to reimbursement and increased testing product numbers; therefore, the data regarding the impact of these two changes is not yet clear. In many European countries, such NGS panels are reimbursed or free at the point of care, such as in the publicly funded UK National Health Service (NHS). This review lacked data from countries outside of the US, which is both key in our interpretation of the results, with US reimbursement most relevant, and also highlights an area of unmet need in research as data begins to emerge from other healthcare settings.
There was noted to be decreased receipt of testing in patients with lung cancer who were current or former smokers as opposed to never smokers [47]. The reasons for this may be linked to increased smoking rates in lower socio‐economic groups [59] or indeed biases interlinked with smoking causation of lung cancer and attached stigma. Furthermore, it has been recognised that non‐smokers are significantly more likely to demonstrate actionable mutations; therefore, clinicians may again prioritise genomic testing in this cohort [60]. Irrespective of rationale, smoking has been surmised to be a barrier to equitable lung cancer care [61].
In the development of the next generation of precision oncology tools, intrinsic socio‐demographic inequalities must be considered. The initial development of breast cancer Oncotype DX utilised information from cohorts in which only 5%–6% of the patients were Black females [62]. This may in part account for discrepancies in the prognostic value of the test, with decreased accuracy in Black cohorts [63]. Certain novel strategies in cancer research, such as machine learning models and other artificial intelligence tools, are particularly sensitive to exaggerating disparities between groups if trained on stereotyped data with intrinsic bias [64]. Of note, this may be present in supervised machine learning models which are being explored to predict ‘high’ and ‘low’ risk breast cancer Oncotype DX groups without necessitating genomic testing [65, 66]. This further highlights the importance of identifying such inequalities where they occur and the active management of future developments to prevent integration of systemic biases which prevent equity of access to novel cancer advances. As clinician education increases as to the benefits of biomarker testing and precision oncology, it is hoped that equitable access is embedded within the anticipated changes in clinical practice.
This review has many strengths in that it provides a contemporary update since 2018 and the broadening of access to NGS panels for solid tumours on the utilisation of genomic testing by socio‐demographic groups. Testing represents an evolving field in which inequalities have been suspected to exist. Combined, data from over 3,200,000 patients across 7 recent years of publication were considered. However, several limitations do exist both methodological and in terms of the evidence base itself. Firstly, there was heterogeneity of classifications of socio‐demographic factors—such as disparities between ethnicity and race and socio‐economic group measures and terminologies. This created challenges in combining such data meaningfully to draw succinct conclusions. Secondly, studies published between 2018 and 2024 did include data from the mid to late 2000s, predating recent advances in precision oncology. Even when used as comparators, these patients are likely to have had differences in diagnostics and treatment options which may confound results. Furthermore, it is acknowledged that in a number of the presented studies, a mix of earlier genetic testing methodologies such as Sanger sequencing were used. When NGS represented the predominant testing modality as per author agreement these were included. It is recognised that in such older cases, any inequalities of access to testing by socio‐demographic factors represent access to the most modern somatic mutation testing available for those patients at the time. ctDNA testing itself is easier for the patient involved, with a simple blood draw yielding contemporary tumour mutational results. Solid tumour testing requires either an invasive up‐to‐date tumour biopsy or utilisation of older tissue, which is less likely to pass quality testing and yield mutation results or may not demonstrate a complete picture of the mutational landscape of the progressed disease at the time of testing. ORs were used when presented in the studies; however, this was not consistent across studies; therefore, a mix of adjusted and unadjusted ORs, as well as hazard ratios when used by the authors, is present for consideration. A single reviewer completing title and abstract screening may have resulted in human error of study omission; however, this is acceptable by the Cochrane Collaboration [67] and on > 10% review of titles and abstracts author agreement by Kappa coefficient was excellent. Within the National Cancer Database studies [14, 30, 34] included on breast cancer Oncotype DX testing there is likely to have been some overlap of patients included; however, the socio‐demographic factors explored differ sufficiently to have not significantly weighted the synthesis of this review. Certain socio‐economic characteristics (e.g., education, smoking and disability status) were not detailed across many studies and may have added depth to the understanding of existing disparities. Finally, data sets from outside the US, and in a diversity of tumour sites outside breast cancer, were lacking and may give a more comprehensive view of worldwide challenges to equitable testing practices in oncology and the generalisability of this review to a non‐US population.
This review especially highlights the lack of evidence in this field, particularly regarding NGS panels, to comprehensively assess equality of access by socio‐demographic factors. This identifies an important area for future research focus. This should encourage robust utilisation data collection moving forwards internationally in order to mitigate the risk of accentuating disparities in cancer care further. What would also be of particular value are studies in publicly funded healthcare settings where access to care is less dependent on insurance or financial status.

5
Conclusions
There are socio‐demographic inequalities in the utilisation of somatic mutation testing in solid tumours. Utilisation is generally lower in patients who are: non‐white; older; with non‐private insurance; and from less affluent socio‐economic groups. Further characterisation and interventions are required to prevent widening of differences in outcome and integration of further systemic inequalities between socio‐demographic groups.

Author Contributions

Sarah Rae: conceptualization (equal), data curation (equal), formal analysis (equal), writing – original draft (lead), writing – review and editing (lead). Annie Baldwin: data curation (equal), formal analysis (equal). Maria Julia Lagonera: data curation (equal), formal analysis (equal). Ruth Norris: conceptualization (equal), data curation (equal), supervision (supporting), writing – review and editing (equal). Alastair Greystoke: conceptualization (equal), data curation (equal), formal analysis (equal), supervision (lead), writing – review and editing (equal). Linda Sharp: conceptualization (equal), data curation (equal), formal analysis (equal), supervision (lead), writing – review and editing (equal).

Funding
S.R. receives fellowship funding from the National Institute for Health and Care Research (NIHR). L.S. and R.N. are funded by the NIHR Newcastle Patient Safety Research Collaboration (PSRC). The views expressed are those of the authors and not necessarily those of the NIHR or Department of Health and Social Care.

Conflicts of Interest
A.G. has received consultancy and speaker fees from Guardant and Foundation Medicine, and is the Clinical Director for Cancer for the North East England and Yorkshire Genomic Medicine Service. The views expressed are those of the authors and not necessarily those of the North East England and Yorkshire Genomic Medicine Service.

Supporting information

Data S1: cam471668‐sup‐0001‐Supinfo.docx.