Current Surveillance Strategy Is Less Effective for Detecting Early-Stage Hepatocellular Carcinoma in Patients with Non-Viral and Non-Cirrhotic Liver Disease Open Access

Hepatocellular carcinoma (HCC) is the third leading cause of cancer-related death worldwide [1] and occurs mostly in patients with established cirrhosis from causes such as alcohol-related liver disease (ArLD), chronic hepatitis B virus (HBV) and C virus (HCV) infections, and non-alcoholic fatty liver disease (NAFLD, recently re-named metabolic dysfunction-associated steatotic liver disease) [2]. At HCC diagnosis, the Barcelona Clinic Liver Cancer (BCLC) staging system stratifies patients according to tumour burden, cancer-related symptoms, and underlying liver function to predict prognosis [3]. HCC is potentially curable if presentation is at an early stage (typically BCLC stage 0 or A), with overall 4-year survival at 47.3% in this group in the UK [4]. Two historical randomised trials in patients with chronic HBV infection and a number of observational studies in patients with cirrhosis showed increased early-stage presentation and improved HCC survival in patients undergoing biannual ultrasound surveillance [5‒8]. As a result, European and American society guidelines recommended that patients with cirrhosis, and selected groups with HBV infection irrespective of fibrosis stage, should undergo 6-monthly ultrasound scan (USS) surveillance with the aim of detecting early-stage HCC amenable to curative therapies and improving overall survival [9, 10].

It is not known whether the benefits of surveillance are the same in different patient groups who may have varied patient- and disease-related risk factors. Much of the observational evidence comes from studies in HBV and HCV, and it is not clear whether these findings can be extrapolated to aetiologies including ArLD and NAFLD. A pooled analysis of patients enrolled in surveillance programmes found lower survival in ArLD-HCC compared with viral-associated HCC, although after adjustments for tumour size, Child-Pugh score, and performance status, the differences were not significant [11], and the study did not evaluate NAFLD. In a large US cohort, there was higher mortality in ArLD-HCC and metabolic disorder-associated HCC, compared with HBV-HCC, after adjusting for demographics, tumour characteristics, co-morbidities, and treatment [12]. Potential reasons for poorer prognosis of ArLD-HCC include lower rates of surveillance in this population and interactions between alcohol toxicity and other risk factors for HCC including diabetes, smoking, and viral hepatitis [13, 14].

NAFLD is the fastest growing cause of HCC in the UK [15], whereas viral-associated HCC is declining largely due to effective anti-viral therapies [15, 16]. The literature provides little specific information for or against surveillance in NAFLD [17] with two small studies, which reported conflicting findings. One found an association between surveillance and early-stage HCC detection in NAFLD (69.6% vs. 30%) [18]; in the other, there was no association between surveillance and receipt of curative treatment or mortality [19]. Suggested explanations for reduced efficacy in NAFLD include increased liver nodularity, making it harder to detect early-stage HCC on ultrasound; higher prevalence of obesity, which can impair performance of ultrasound; and increased risk of liver and non-liver mortality in NAFLD cirrhosis mitigating surveillance-related survival benefits [17]. There is evidence of ethnic disparity in HCC stage at presentation and survival in the USA [20, 21]; patients of white ethnicity present earlier and survive longer compared to people of Hispanic ethnicity. We and others have previously shown different prevalence rates of progressive liver disease and earlier progression to fibrosis in NAFLD in different ethnic groups [22‒24]. Ethnicity is acknowledged as a risk factor for disease in screening guidelines for people with HBV infection, and surveillance starts earlier in people of Asian or black African ethnicity, but guidance is not tailored to ethnicity in NAFLD.

We hypothesise that stage at presentation and all-cause mortality from HCC are affected by patient- and disease-related factors including liver disease aetiology and ethnicity and that current surveillance practices may not be effective in all aetiologies of liver disease. To test this, we conducted a multicentre retrospective observational study including patients with newly diagnosed HCC from six UK centres.

Adult patients aged 18 years and above, with new diagnosis of HCC, confirmed on radiology and/or histology, and referred to any of six regional HCC multidisciplinary team (MDT) meetings were included in this study. A total of 1,780 patients from six UK centres (London [Barts Health], Oxford, Nottingham, Birmingham, Coventry, Bristol) were included in the study. This comprised all patients referred to Barts Health HCC MDT (n = 817) between January 1, 2007, and December 31, 2020 (cohort 1), and all patients with available data referred to HCC MDT in Oxford, Nottingham, Birmingham, Coventry, and Bristol over the same time period (cohort 2, n = 963). Follow-up data were captured until December 31, 2020, for all centres.

Data collected from MDT outcome forms included date of MDT meeting, date of birth, sex, aetiology of liver disease, Barcelona Clinic Liver Cancer (BCLC) stage, and HCC surveillance. Date of diagnosis was defined as the date of the HCC MDT where the case was reviewed. Missing data were retrospectively collected from patient records according to pre-defined criteria, where available. When collected retrospectively, HCC surveillance was defined as a screening liver USS within the last 12 months before MDT date, to reflect the real-world nature of this study and allow for delays due to patient- and system-related factors, such as rescheduled or delayed USS appointments or delay in referral to MDT. Aetiology of liver disease was defined as the primary diagnosis in hepatology clinic letter or, where not available, coded liver disease diagnosis. BCLC stage was determined from the healthcare record according to tumour burden, liver function (by Child-Pugh stage), and performance status. Also collected from patient records were coded patient-reported ethnicity; type 2 diabetes mellitus (T2DM) diagnosis; metformin use at diagnosis (defined as prescribed metformin for 6 months prior to diagnosis, where data available); HbA1C (within 6 months of diagnosis), cirrhosis at time of diagnosis (defined as ultrasound/CT/MRI evidence of cirrhosis, cirrhosis on liver biopsy histopathology report, MDT documentation of cirrhosis, or documentation of cirrhosis in clinic letter by consultant hepatologist), and outcome at the end of the follow-up period (alive, date of death, or date lost to follow-up).

Aetiology of liver disease was characterised as ArLD, HBV, HCV, NAFLD, and all other aetiologies (including diagnosis coded as “cryptogenic” and “none”) based on either electronic healthcare record coding or documentation on HCC MDT proforma/outpatient clinic letters. Where multiple aetiologies were present, patients were classified according to first diagnosis or most significant diagnosis as per hepatology clinic letter or MDT. In this manuscript, we use the term “NAFLD” rather than “metabolic dysfunction-associated steatotic liver disease” as per the clinical records. We have not substituted the new term in our analyses as patient co-morbidities were not extensively documented to ensure the presence of a metabolic risk factor.

BCLC stage 0 and A disease at presentation was defined as early-stage disease. BCLC stage C and D at presentation were defined as late-stage disease. Ethnicity was collapsed into five categories: black, East Asian, South Asian, white, and other. The study was approved by each centre’s Clinical Standards and Audit Department or equivalent as a service evaluation of outcomes in HCC and therefore individual informed consent was not required or taken.

We used Cox proportional-hazards regression models to examine the relation between aetiology of liver disease, surveillance, and ethnicity with all-cause mortality. Associations were summarised with adjusted hazard ratios (aHRs) and 95% confidence intervals.

We used multivariable logistic regression models to examine the relation between aetiology of liver disease, surveillance, and ethnicity with HCC stage at presentation. Associations were summarised with adjusted odds ratios (aORs) and 95% confidence intervals.

All regression models were adjusted for age, sex, aetiology of liver disease, and surveillance. Cox proportional-hazards models were also adjusted for cirrhosis. Multivariable logistic regression models examining HCC BCLC stage at presentation were not adjusted for cirrhosis as liver function is a component of BCLC staging. All models were adjusted for centre to account for between-centre differences.

We performed subset analysis in patients with ArLD, viral hepatitis (HBV and HCV combined), and NAFLD to examine whether surveillance-associated outcomes are altered by underlying liver disease aetiology. We also performed subset analysis in patients with non-cirrhotic HCC, T2DM, and in those with T2DM who were taking metformin.

Age and HbA1C were treated as continuous variables. Sex, surveillance, T2DM, and metformin use were treated as binary variables. Aetiology was treated as a categorical variable, and patients were categorised as having one aetiology only. In the regression models, ArLD, HBV, and HCV were the explanatory variables and NAFLD and all other aetiologies were the reference category, given the degree of overlap between them due to under-diagnosis of NAFLD. Given the relatively small number of patients in most ethnicity categories, in the regression models, ethnicity was treated as a categorical variable where white ethnicity was the explanatory variable and all other ethnicities were the reference category. In cohort 2, centre was treated as a categorical variable with Birmingham as the reference category.

Categorical variables were compared using the chi-squared test. We used SPSS software, version 28.0 (IBM), for statistical analysis. Two-tailed p values of <0.05 were considered to indicate statistical significance.

A total of 1,780 patients with HCC, including 354 with ArLD, 200 HBV, 325 HCV, and 503 NAFLD, were included for analysis (Fig. 1a). The median age was 68 years (IQR: 60–76) and 80.2% were male (Fig. 1b). The commonest ethnicities were white (70.1%), followed by South Asian (15.9%) and black (7.3%) (Fig. 1c). NAFLD was the commonest aetiology in white and South Asian ethnicities (34.3% and 34.0%, respectively), whilst viral hepatitis accounted for majority of cases in East Asian and black ethnicities (83.0% and 77.1%, respectively; online suppl. Table 1; for all online suppl. material, see https://doi.org/10.1159/000542805; Fig. 2). Overall, 78.2% of patients had documented evidence of cirrhosis (59.6% Child-Pugh A, 26.2% Child-Pugh B, and 14.3% Child-Pugh C cirrhosis; Fig. 1d) with the highest proportion amongst those of white ethnicity (78.5%) and lowest in those of black ethnicity (66.1%; online suppl. Table 1). Only 43.0% of patients were under surveillance for HCC at the time of diagnosis (Fig. 1e) with the highest proportion amongst those of white ethnicity (43.2%) and lowest in those of East Asian ethnicity (31.9%). The most common BCLC stage at time of diagnosis was A (39.0%), followed by C (20.9%), B (19.5%), D (17.5%), and 0 (3.1%) (Fig. 1f). BCLC stages C and D, which represent advanced/terminal stage of HCC, were highest in those of black ethnicity (51.7%) and lowest in white ethnicity (36.8%) at the time of HCC diagnosis (Table 1).

Patients from cohort 1 (single centre, n = 817) were diagnosed between 2007 and 2020. In multicentre cohort 2 (n = 963), the majority of patients were diagnosed between 2017 and 2021 (online suppl. Fig. 1). Of the 817 patients in cohort 1, 55.1% were of white, 25.8% South Asian, 12.1% black, and 4.9% East Asian ethnicities (Table 1). Cohort 2 was less ethnically varied, with white patients comprising 84.7% (Table 1). The most common aetiology of liver disease in cohort 1 was HCV (27.1%), followed by NAFLD (25.7%), whereas the most common aetiology in cohort 2 was NAFLD (33.5%), followed by ArLD (26.7%). In both cohorts, the proportion of patients with non-cirrhotic HCC was highest in patients with NAFLD in whom it was more than the combined proportion of patients with HBV, HCV, and ArLD (cohort 1: 32% vs. 13%, p < 0.001; cohort 2: 16% vs. 6%, p < 0.001; Fig. 2). A total of 564 (69.0%) patients died and 108 (13.2%) were lost to follow-up in cohort 1 compared to 474 (49.2%) and 223 (23.2%), respectively, in cohort 2.

We determined the relationship of patient- and disease-related factors with all-cause mortality and early-stage disease at presentation using Cox proportional-hazards and logistic regression models. Overall, older age was associated with higher all-cause mortality and lower likelihood of presenting with early-stage disease after adjustments, but sex did not have any effect (Tables 2, 3). This was observed in both cohorts 1 and 2 when analysed separately. White ethnicity was independently associated with lower all-cause mortality (aHR: 0.80, p < 0.01) and higher likelihood of presenting with early-stage disease (aOR: 1.44, p = 0.02) (online suppl. Table 2a, b). Cohort 2 was not sufficiently ethnically diverse to include in this analysis. As expected, cirrhosis was independently associated with higher all-cause mortality (Table 2). The aHR for cirrhosis was higher than other patient- or disease-related factors.

T2DM is an important risk factor for the development of HCC and cirrhosis [25] and was prevalent in 44% of patients. The majority of patients with T2DM had NAFLD (55.7% overall, 51.4% in cohort 1, and 59.4% in cohort 2). Overall, we did not find a statistically significant effect of T2DM on all-cause mortality (aHR: 0.91, p = 0.24; online suppl. Table 3), although this was not consistent in both cohorts (cohort 1: aHR: 1.02, p = 0.88; cohort 2: 0.77, p = 0.03). However, in patients with T2DM who were prescribed metformin, we found evidence of benefit. Metformin prescription was independently associated with lower all-cause mortality overall and in cohort 2 (aHR: 0.60, p < 0.01; online suppl. Table 4a) and with early-stage disease at presentation in cohort 1 (aOR: 2.75, p = 0.01; online suppl. Table 4b), which remained significant after adjustment for HbA1c levels (aOR: 3.85, p = 0.023).

HCC surveillance was associated with better survival and earlier stage of disease at presentation. People who underwent HCC surveillance had lower risk of all-cause mortality (overall aHR: 0.67, p < 0.01; cohort 1 aHR: 0.66, p < 0.01; cohort 2 aHR: 0.66, p < 0.01; Table 2) and, consistent with this, were more likely to present with early-stage disease (overall aOR: 2.69, p < 0.01; cohort 1 aOR: 2.63, p < 0.01; cohort 2 aOR: 2.68, p < 0.01; Table 3). To accommodate the real-world nature of this study, a broader window of up to 12 months was allowed to define surveillance. To determine whether this underestimated the benefits of surveillance, we performed sensitivity analysis in patients from cohort 1 with available dates for surveillance scans. We did not find a statistically significant difference in rates of survival or early stage of disease at presentation between people whose surveillance scan was within 6 months of diagnosis compared to those with scan between 6 and 12 months (online suppl. Table 5a, b).

Aetiology of underlying liver disease was associated with all-cause mortality but not early-stage disease at presentation (Tables 2, 3). Overall, HCV was associated with a lower rate of all-cause mortality (aHR: 0.76, p = 0.01). In cohort 1, HCV and HBV were associated with a lower rate of all-cause mortality (aHR: 0.61, p < 0.01, and aHR: 0.72, p = 0.03, respectively), and there was a non-significant trend between ArLD and a higher rate of all-cause mortality in cohort 2 (aHR: 1.24, p = 0.07). To determine whether surveillance-associated outcomes for HCC are altered by the underlying liver disease aetiology, we performed sub-analyses in patients with viral hepatitis (combining HBV and HCV), ArLD, and NAFLD. In patients with viral hepatitis, surveillance was associated with lower all-cause mortality (aHR: 0.46, p < 0.01; Table 4) and greater likelihood of early stage at presentation (aOR: 4.16, p < 0.01; Table 5). In contrast, in patients with ArLD, surveillance was not associated with all-cause mortality but greater likelihood of early-stage at presentation (aOR: 1.92, p = 0.02) (Tables 6, 7). Females with ArLD (but not other aetiologies) were significantly more likely to present with early-stage HCC (aOR: 2.57, p = 0.02; Table 7). Similar to ArLD, in patients with NAFLD, surveillance was independently associated with early stage at presentation (aOR: 2.11, p < 0.01) but not all-cause mortality (Tables 8, 9). Furthermore, fewer patients with NAFLD were undergoing HCC surveillance at presentation, compared with all other aetiologies (overall: 36.0% vs. 47.1%, p < 0.01; cohort 1: 19.8% vs. 37.9%, p < 0.01; cohort 2: 47.8% vs. 57.2%, p = 0.01). This corresponded with the higher number of patients with NAFLD who had non-cirrhotic HCC compared to the combined proportion of patients with HBV, HCV, and ArLD (23% vs. 11%, p < 0.01; Fig. 2). Similar to patients with cirrhosis, younger age, being under surveillance, and HCV were significantly associated with survival in patients with non-cirrhotic HCC (online suppl. Table 6a). Our data did not allow for calculation and correction of lead-time bias at individual patient level. Therefore, we subtracted lead time from overall study time in patients undergoing surveillance [26], using an estimate of lead time available from the literature (6.5 months, derived from Cucchetti et al. [27]), and found that surveillance was not a statistically significant predictor of overall survival (aHR: 0.87, p = 0.07; online suppl. Table 7a), but it was associated with survival in patients with viral hepatitis (aHR: 0.57, p < 0.01; online suppl. Table 7b), further supporting our findings that current HCC surveillance is less effective in non-viral liver disease. Taken together, we found that HCC surveillance was associated with improved survival and early stage at presentation in viral hepatitis but not ArLD or NAFLD.

In this multicentre study of HCC-associated outcomes in the UK, we found that HCC surveillance was associated with survival and early-stage disease at presentation in patients with viral hepatitis in line with existing literature [28] but was less beneficial for patients with non-viral aetiologies (ArLD and NAFLD). Less than half of patients diagnosed with HCC had undergone surveillance, including many without cirrhosis who fall outside current surveillance guidelines. We identified patient- and disease-related factors that independently affect HCC outcomes. Females were more likely to present with early-stage ArLD-HCC compared to males and patients of white ethnicity had a survival benefit and were less likely to present with late-stage HCC compared to other ethnic groups.

Our results reflect the changing demography and epidemiology of chronic liver disease with rising numbers of patients with NAFLD as the underlying aetiology. In our study, one-third of NAFLD-HCC patients did not have cirrhosis and were not under surveillance for HCC, consistent with current guidance and with existing literature; 34.6% of NAFLD-HCC patients were not cirrhotic in the US Veterans Administration series [29]; and, in meta-analyses, 38–39% of NASH-HCC patients were not cirrhotic compared with 14.2–14.6% for all other aetiologies [30, 31]. In patients with non-cirrhotic HCC, we identified that younger age and HCV were associated with survival. Current HCC surveillance programmes focus on patients with cirrhosis and may require modification to include pre-cirrhotic NAFLD patients with risk factors for HCC such as T2DM in those with advanced fibrosis as suggested by some organisations [32] and refinement with biomarkers or genetic risk stratifiers [33]. The current bi-annual surveillance strategy with ultrasound evolved from a randomised controlled trial in patients with HBV from China [6]. It is unclear whether this approach is as effective as in people with NAFLD or ArLD, given differences in body habitus and ultrasound technical success, heterogenous liver echotexture, and lesion recognition. Ultrasound sensitivity for detecting HCC was 77% in patients with BMI <30 kg/m² reducing to 21% in patients with BMI ≥30 kg/m² [34]. In a study of 941 patients with cirrhosis, obesity, non-viral aetiologies of liver disease, and Child-Pugh B cirrhosis were all independently associated with worse ultrasound quality for evaluation of HCC [35]. In a recent pooled meta-analysis of 59 studies in HCC surveillance, only two examined the association between surveillance and clinical outcomes among those with NAFLD and reported opposing survival outcomes [28]. HCC surveillance was associated with improved survival in studies where NAFLD accounted for <20% of total HCC cases [28]. Survival in studies containing >20% NAFLD-associated HCC is currently unclear due to paucity of data, which is complemented by our study in which NAFLD was the largest underlying liver aetiology (30%). Alternative imaging-based surveillance tools such as MRI [36] and biomarker-based surveillance tests such as GALAD score [37] have been developed, which have demonstrated improved sensitivity for HCC detection compared with existing surveillance strategy. Further studies are needed to assess the best surveillance tool and strategy, particularly for non-viral and non-cirrhotic HCC.

Given the real-world nature of our study, our study design was intended to reflect clinical practice, where a strict 6 months window for ultrasound surveillance is not always possible for intended bi-annual ultrasound scans due to patient-related and healthcare provider-related factors, hence considering patients with a scan in the 12 months preceding diagnosis in the surveillance group.

T2DM is a key risk factor for liver disease progression [25, 38] including towards HCC, but data regarding T2DM as an independent predictor for outcomes are more equivocal [39‒41]. Some data suggest that T2DM is associated with poorer prognosis in BCLC stages 0, A, and B but not in those with BCLC stage C and D disease [42] for whom the cancer burden may eclipse the T2DM effect. While we found that T2DM had no significant impact on survival in HCC in cohort 1, it was associated with a survival benefit in cohort 2 (after adjustments including aetiology and cirrhosis), which may reflect a benefit from being prescribed medications used in T2DM, including metformin. Metformin has been associated with reduced risk of developing HCC [43, 44], with improved overall survival [41, 45] and with specific survival benefit in patients undergoing TACE and radioablation [46]. Although we found benefit related to metformin use, the protection was inconsistent between the two cohorts. Reasons for this may include differences in cohorts 1 and 2 in terms of data availability, patient demographics, and liver disease aetiology and changes in diabetes care over the time period of this study (2007–2021). We did not capture use of other anti-diabetic medications such as GLP-1 agonists shown to improve all-cause mortality in people with diabetes [47]. It is not clear in the current study whether metformin use in the real world is a surrogate for engagement with healthcare, health behaviours or whether there is genuine biological benefit. Our study provides rationale for a larger study to understand the risks associated with T2DM and potentially prospective randomised studies to assess whether metformin use can mitigate against these in HCC.

Although females are more suspectable to alcohol-induced liver injury than males globally [48], age-standardised death rates from ArLD-associated HCC have increased in males and decreased in females over the last decade [49]. This is in line with our findings that females with ArLD were 2.5-fold more likely to present with early-stage HCC compared to males.

Studies from the USA report ethnic disparity in HCC stage at presentation and survival [50‒52]; compared to Hispanic and black patients, white patients presented earlier and survived longer. We observe similar findings in the current study: an ethnically diverse UK cohort where patients of white ethnicity were less likely to present with late-stage HCC and were more likely to survive, compared to other ethnicities. Further work is needed to determine patient- and disease-related factors and to address them in order to reduce existing disparities.

Our study has several limitations. Its retrospective nature is likely subject to lead-time bias, and so associations may be overestimated. However in line with our findings, a recent UK prospective study, which adjusted for lead time, found that surveillance was associated with improved overall HCC survival [53]. Our data were obtained by review of real-world patient records, performed retrospectively, which precluded verification. We were unable to capture the indication for the diagnostic tests beyond the surveillance programme. One might speculate that patients who did not meet our definition for surveillance may still have been in care and attending for diagnostic investigations, while others may have had lesions identified incidentally during testing for other indications including abnormal liver biochemistry or abdominal symptoms. Our study characterised patients as having a single aetiology of liver disease, whereas in reality, patients may have mixed aetiologies. All HCC diagnoses were diagnosed through referral to regional HCC multi-disciplinary meetings, which may be subject to referral bias. However, the regional nature of the multi-disciplinary meetings is such that the patients we identified have been referred from different centres across the region and not only the tertiary centres themselves. Viral control status, such as sustained virological response in chronic hepatitis C and undetectable HBV DNA in chronic hepatitis B, was not collected in this study. Finally, data on tumour markers, treatments, and responses following HCC diagnosis were not captured in this study.

In this study, we identified that less than half of patients presenting with HCC had undergone surveillance. HCC surveillance was associated with improved survival and early stage at presentation in viral hepatitis but not ArLD or NAFLD suggesting that the existing surveillance strategy is less effective for detecting early-stage HCC in patients with non-viral aetiologies. Social determinants such as ethnicity are relevant. Given the growing prevalence of NAFLD and metabolic syndrome, better understanding is required of the roles of T2DM and metformin in HCC outcomes. HCC surveillance strategies that benefit this patient group, and in particular those without cirrhosis, are needed.

The study protocol was reviewed and approved by each centre’s Clinical Standards and Audit Department or equivalent as a service evaluation of outcomes in HCC and therefore ethics approval and individual informed consent were not required. Participating site Clinical Standards and Audit Department information can be found at https://www.jrmo.org.uk/performing-research/conducting-research-in-the-nhs/setting-up-a-study/, https://www.nuh.nhs.uk/approvals-research/, https://www.ouh.nhs.uk/researchers/planning/is-it-research/https://www.research.uhb.nhs.uk/research-and-innovation-at-uhb/information-for-researchers/, https://www.uhcw.nhs.uk/our-organisation/publications-and-documents/, https://www.uhbristol.nhs.uk/for-clinicians/clinicalaudit/.

W.A. has received honoraria and for speaking and consultancy from GlaxoSmithKline, Gilead Sciences, UCB Biopharma, and Conclusion, is a Scientific Advisor for Metadeq Inc., and is in receipt of competitive funding from GlaxoSmithKline, MSD, and Gilead Sciences. S.S. receives honoraria from Eisai for speaking and consultancy from Faron Pharmaceuticals. E.L.C. receives funding from NIHR BRC Oxford. Other authors had nothing to declare.

There was no funding of any research relevant to this study.

Concept and design: Wenhao Li, Jessica Spiers, and William Alazawi. Data acquisition and analysis: Wenhao Li, Jessica Spiers, Hamish Miller, Rosemary E. Faulkes, Muhammad Saad, Abhishek Sheth, Katharine Caddick, Yaqza Hussain, Victoria Gordon, Jenny Merry, Mohsan Subhani, Muhammad Nauman Tahir, and Esther Unitt. Initial drafting of manuscript and equal contribution to manuscript: Jessica Spiers and Wenhao Li. Critical review and adaptation of the manuscript: Wenhao Li, William Alazawi, Aloysious D. Aravinthan, Ayman Bannaga, Tahir Shah, Emma L. Culver, Shishir Shetty, Ankur Srivastava, and Nwe Ni Than. All authors approved the final version.

Jessica Spiers and Wenhao Li: joint first authors.

The data that support the findings of this study are not publicly available because they contain information that could compromise the privacy of participants but are available from Wenhao Li and corresponding author William Alazawi upon reasonable request.