The ECHO study is a cross-sectional investigation conducted amongst persons without stable housing in France. The findings presented here are based on two waves of data collection, taken during the Spring of 2020 (02/05/20–07/06/20) (W1) and 2021 (09/03/21–31/05/21) (W2). Whilst the samples for each wave were distinct, the methodology was generally consistent between the two. Participants were recruited from shelters located in the regions of Paris (W1 n = 12; W2 = 9), Lyon (W1 = 5; W2 = 10) and Strasbourg (W1 = 1). Recruitment for W2 also incorporated day centres (n = 7), allowing the inclusion of those sleeping rough. Interviews were conducted both in person (W1 = 98%; W2 = 100%) or by telephone (W1 = 2%), in French, English or participants’ chosen language, with the help of independent interpreters, to minimise bias due to language barriers (W1 = 33%, W2 = 13%). Over 20 languages were used, most frequently Arabic, Pashto and Dari. Participants were excluded if aged under 18 years, significantly inebriated or presenting cognitive disorders that prevented consent. The study protocol was approved by the Ethical Research Committee of the University of Paris (CER-2020-41).

The assessment of both substance use and risk factors was done via direct, one-on-one interviews with participants. All questions were based on a pre-determined, quantitative questionnaire, designed in each study wave by the ECHO research team. Whilst the questionnaire could have been self-administered, interviews were conducted to reduce exclusion based on participants’ language or literacy. In both study waves, participants were asked to report their current use of tobacco and alcohol on a 3-point scale ranging from “every day,” “sometimes”; or “no”. For multivariate analyses, “every day” and “sometimes” were combined, generating two categories for current use (yes/no). As a secondary objective of this study, participants also reported cannabis use on the same scale, however, due to its illegality, a “does not wish to answer” option was available in W1. For W2, questions on substance use were exclusive to the supplementary questionnaire, and refusal to answer was registered as missing data.

Factors considered potentially relevant to patterns of substance use were the following: age (18–29; 30–49; 50 + years), sex (male; female), partnership status (stable partner; single), family status (no children; living with children; has children but living separately), educational level (no school/incomplete primary education; primary/high school education; College/higher education), employment (none; only before lockdown; both before and during lockdown), duration of stay in France (<1 year; 1–5 years; 5+ years, including French natives), French language aptitude (low; moderate/fluent), administrative status (French native; residence permit holder; asylum seeker; no residence permit; other), health insurance (yes; no), chronic illness (yes; no), food insecurity (yes; no), feelings of safety (yes; no), exposure to theft or assault during the first lockdown (yes; no), contact with friends/family (yes; no), and participants’ previous accommodation (centre/association; unestablished shelter e.g., camps, squats; street; friends/family/other).

French language aptitude was calculated from the sum score of self-reported French speaking, reading and writing ability, each rated on a 4-point Likert scale. Participants’ degree of loneliness was measured based on the UCLA loneliness scale [30]. Health literacy was calculated as a sum of participants’ scores to either the fifth (W1) or fourth and seventh (W2) dimensions of the Health Literacy Questionnaire [31], with the threshold between low and high set as an average agreement of ≤50% for W1 and ≤33% for W2. Depression was assessed via the nine-item Patient Health Questionnaire (PHQ-9) in W1 (cut-off score 10 [32]), and the four-item PHQ-4 in W2 (cut-off score 5). Both are validated for use in multicultural settings [33, 34]. Finally, participants region of origin was divided into France; Europe excluding France; the Middle East; North Africa; Sub-Saharan Africa; and Other (categories defined in Supplementary Table S1).

To identify factors associated with alcohol and tobacco use, the following approach was implemented. First, differences in the prevalence of substance use between Spring 2020 and 2021 were examined using Chi-square (X²) analysis. We then tested associations with factors considered relevant a priori, in relation to substance use, using generalised logistic regression models, implementing separate models for each study wave. Missing covariate values were imputed using Multivariate Imputation by Chained Equations (MICE) [35].

Participants from American, South-East Asian, and Western Pacific regions were excluded due to insufficient sample size (W1 n = 5; W2 = 12). Variables included in multivariate statistical models were determined via univariate X² analyses, as supported by Hosmer and Lemeshow [36, 37]. An 80% confidence limit was used [38], to prevent the arbitrary exclusion of important variables [39, 40]. For measures of prevalence, cannabis use was also included, allowing us to further investigate interactions between respective substances.

All data analyses were performed on R Version 4.1.1 [41, 42].

Overall, 1564 persons were invited to take part in the ECHO study. Following participant refusal or unavailability, 535 participants were interviewed in Spring 2020, of which 530 had sufficient data on substance use, and 523 in Spring 2021, of which 319 had sufficient data on substance use. Demographic information on both study samples is available in Table 1.

Rates of substance use are shown in Table 2. Tobacco was the most commonly used substance in both study waves (42% of participants in Spring 2020; 38% in 2021), followed by alcohol (26%; 34%), and cannabis (10%; 9%). Univariate analyses found tobacco use to be significantly associated with the use of both alcohol (p < 0.001) and cannabis (p < 0.001) in both waves. Alcohol use also significantly associated with cannabis use (p < 0.001).

Differences between rates of substance use between the two study waves are shown in Figure 1. When compared to 2020, alcohol use was more prevalent in 2021 (p < 0.001).

Factors associated with alcohol and tobacco use, determined via multivariate regression, are shown in Table 3. In Spring 2020, characteristics associated with tobacco use were being male (aOR:5.04; 95% CI: 2.43–10.46), not having a stable partner (aOR:2.01; 95% CI: 1.16–3.46), being medically uninsured (aOR:2.10; 95% CI: 1.21–3.64) and exposure to theft or assault (aOR:2.55; 95% CI: 1.32–4.93). Compared to French natives, participants born in the Middle East (aOR:0.43; 95% CI: 0.21–0.88) or Sub-Saharan Africa (aOR:0.22; 95% CI: 0.11–0.41) were less likely to smoke tobacco, whilst those born in other European countries were more likely (aOR:4.54; 95% CI: 2.03–10.14). Participants who had lived in France for more than 5 years were also more likely to smoke than those living in France for less than 1 year (aOR:2.34; 95% CI: 1.03–5.31). In Spring 2021, tobacco use associated with being male (aOR:4.42; 95% CI: 2.09–9.37) and medically uninsured (aOR:2.88; 95% CI: 1.09–7.64).

Characteristics associated with alcohol use in Spring 2020 were being male (aOR:3.44; 95% CI: 1.63–7.25), exposure to theft or assault (aOR:1.83; 95% CI: 1.01–3.31) and experiencing symptoms of depression (aOR: .66; 95% CI: 1.00–2.75). In Spring 2021, alcohol use associated with being male (aOR:2.53; 95% CI: 1.26–5.07) and having had primary or high school education (aOR:3.80; 95% CI: 1.44–10.00) as opposed to no schooling. Compared to French natives, North African participants were less likely to drink alcohol (aOR:0.24; 95% CI: 0.11–0.55).

Tobacco was the most frequent substance used in both ECHO study waves. Cross-sectional data on the French general population during Spring 2020 saw 21% were tobacco smokers (occasional and regular) [4]. Our findings then suggest a rate amongst homeless persons roughly double that of the general population. Considering that homeless persons present higher rates of medical comorbidities [25, 26, 45] and are therefore more at risk of tobacco-related illness or fatality [26, 46], these findings are cause for concern. Additionally, despite universal health care, socioeconomically disadvantaged groups are less likely to have access to smoking cessation programs, leading to large inequalities in smoking which appear to have increased in the context of the COVID-19 pandemic [47]. Regarding alcohol, rates within ECHO were also higher than national levels (17% regular drinkers) [48].

Whilst the rates of tobacco use were similar between Spring 2020 and 2021 (38%–43%), alcohol use was more frequent amongst participants in 2021 (34%) than 2020 (26%). Several key factors may account for the higher rate of alcohol use seen in Spring 2021; first, during the initial lockdown period, restrictions may have decreased the availability of alcohol, both physically (bar/shop closures; exit permits) and financially (reduced hours; job loss) [17]. Many of these restrictions were either lifted or relaxed by Spring 2021, during which France experienced its third lockdown. This can be seen, for example, in the rate of employment. Among the ECHO population surveyed in Spring 2020, 93% were unemployed, compared to just 79% in Spring 2021. Whilst employment status did not associate with alcohol use in our study population, nor amongst non-homeless populations during the pandemic [49], factors associated with fewer restrictions (such as increased socialisation and disposable income) may have enabled more frequent drinking [50]. The higher rates of alcohol use amongst those surveyed in 2021 may also be specific to each groups’ respective living situation. All participants in the first wave were housed in temporary shelter, which may have had a positive influence on alcohol consumption. A stable, communal living environment can provide greater support than is often available for those sleeping rough, alongside that from social workers specifically hired by the organisations themselves [51]. On-site case-management and social support, given through the Housing First initiative, has been shown to reduce alcohol-use amongst previously-unsheltered homeless persons [52, 53]. Furthermore, qualitative findings collected during April–August 2020 show homeless-centre users found support services acted as a ‘lifeline’ during the pandemic, providing invaluable stability during a time of such disruption [54]. In comparison, the ECHO sample interviewed in Spring 2021 had a variety of living situations. Alongside a reduced level of support, this may elevate rates of alcohol use bidirectionally: those who drink may struggle more to find or maintain shelter, whilst those without temporary shelter may then be more likely to drink [55] Moreover, the discontinuation of the housing services initially provided in response to the pandemic (of which the first-wave participants benefitted) may have triggered alcohol consumption. In support of this, Scallan et al. [29] found disruptions to emergency housing, provided temporarily during the pandemic, resulted in increased addictive substance use. Together, our findings may then indicate the importance of a consistent, supportive living environment for the promotion of healthy behaviours, particularly following the fluctuations in housing caused by the pandemic.

Several of these factors, however, would also apply to tobacco, despite no significant difference being noted. Our findings also contrast with those from the general French population, which found tobacco smokers were more likely to report an increase in use (27%) since the onset of the pandemic, whilst alcohol use generally decreased—at least in frequency [52]. This may then indicate that alcohol drinking was a more popular coping mechanism for the progressive, long-term strain of the pandemic amongst homeless populations [54]. In support of this, drinking to cope during the pandemic was seen to associate with depression [49], for which the ECHO population showed higher-than-national levels [55]. Alcohol use within both study populations may have also depended more on social liberty (the ability to socialise; beg for money) and financial security than tobacco. In fact, several studies showed social isolation was not a significant risk factor for changing tobacco consumption during the pandemic, with one even finding it to be the most popular reason for decreasing cigarette use [56, 57].

These results may be useful to guide both the prevention and treatment of substance abuse in homeless populations following the COVID-19 pandemic. Existing preventative measures for addictive behaviour might not be effective; studies have shown that whilst homeless adults who have experienced more frequent homeless episodes have higher odds of receiving anti-smoking support, they show lower rates of quitting [26]. Moreover, several of the risk factors identified, such as exposure to assault and symptoms of depression, have also increased in prevalence during the pandemic [66, 67, 87, 88].

Whilst risk factors associated with substance use varied between tobacco and alcohol, the use of different substances strongly associated with each other. Research suggests many of the genetic and environmental vulnerabilities towards using different addictive substances are shared [89, 90], therefore the risk factors identified for tobacco and alcohol use within this sample may remain pertinent to other substances. Considering the concurrent use of multiple substances also increases the chance of subsequent injury [91], interventions considering all potential risk factors may be more successful in preventing substance-related morbidity and mortality.

Our study has several limitations which need to be addressed. First, the primary focus of the ECHO study was not to investigate substance use. Therefore, our assessment was relatively limited, comprising of one multiple-choice measure per substance. Despite this, the more generalised questionnaire design allowed us to account for a wider range of variables in the limited interview time available, often not examined in relation to substance use. Secondly, ECHO comprises two cross-sectional waves of data collection, based on separate samples. This therefore limits our assessment of longitudinal patterns of substance use. The most significant difference between the two study waves was participants’ housing situation; whilst wave one included only persons living in temporary accommodation at the time of investigation, wave two recruited persons from a wider range of situations. However, it is important to note that during the Spring of 2020, France had an active policy of providing temporary shelter to all persons sleeping rough, in order to limit the propagation of COVID-19. This sample may thus be more reflective of the general homeless population than those typically residing in temporary accommodation. In support of this, 41% of wave one participants were living on the street before their current shelter, thereby indicating the inclusion of various experiences of homelessness.

Finally, this study may be biased due to our reliance on participants’ self-reports. In comparison to biological markers, self-reports are known to generate underestimations of substance use [43]. Moreover, within our study, both data collection points occurred during Ramadan, a month associated with abstinence from drugs within the Muslim community. Although no data on participants’ religious practice were collected, 61% of our sample originated from a predominantly Muslim country. If practising Islam, these participants are then potentially more likely to have both reduced their consumption during this period, and under-report their levels of use due to social or cultural desirability bias [44]. This tendency towards underestimation may have also been accentuated by the precariousness of our study population’s accommodation, as participants may have feared that reporting recreational drug use would have jeopardised their right to shelter. To counteract this, before every interview, participants were reminded that their responses were anonymous and would not impact their right to accommodation, and that they were free to refuse any question.

Nevertheless, our study has numerous strengths which support the validity of our findings. The data available on substance use within homeless populations is limited, and whilst our findings cannot be extrapolated to pre- or post-pandemic periods, factors associated with substance use may remain relevant. Moreover, the inclusion of multiple different substances within the same sample provides a particularly valuable, often-unreported comparison between substances with consistent demographics. Our study was conducted in two large metropolitan regions of France, thereby limiting the role of specific contextual factors on the selection of the homeless population living in accommodation centres. Finally, we interviewed participants who could not speak French or English, through professional interpreters, making it possible to collect data among recent migrants who constitute the largest share of the current homeless population in France.

Our study shows elevated rates of substance use among homeless persons during the COVID-19 pandemic. Tobacco was the most commonly used substance, with rates roughly double that of the general population. The use of different substances also positively associated with each other, with the majority of users using more than one substance. Factors associated with tobacco and alcohol use varied, except being male and exposed to theft or assault, which increased the risk of using both substances. Higher rates of alcohol use were seen in the later stage of the pandemic, which may associate with participants’ greater housing insecurity during this wave. Together these findings provide valuable information on those most vulnerable to substance abuse within homeless populations. Considering the increased risk for both alcohol and tobacco-related injury amongst vulnerable persons, adapting interventions to better prevent substance use amongst those experiencing homelessness in the context of the COVID-19 pandemic is a public health necessity.