无家可归者的死亡率：一项系统综述和荟萃分析。
全文（开放获取） · 中文译文
摘要 背景：无家可归以及无家可归经历类型与全因及特定原因死亡率之间的关联尚不确定。 方法：通过系统检索MEDLINE、Embase PsycINFO和Scopus自建库至2024年12月的已发表研究，并通过开放获取数据档案确定未发表数据。我们使用随机效应Meta分析合并来自116项研究的已发表和未发表数据的效应估计值，这些研究包括2 563 633名无家可归者和129 292 553名人群对照。本综述已在PROSPERO注册（CRD42023430984）。 结果：与未暴露者相比，暴露于无家可归者的全因死亡风险增加[相对风险（RR）2.21，95%置信区间（CI）1.91-2.57，P < .001，I2 = 99.7]，男性和女性风险相似（男性RR 4.04，95% CI 2.87-5.21；女性RR 3.58，95% CI 2.35-4.82）。露宿者的风险（RR 7.63，95% CI 3.29-11.97）高于在无家可归者招待所/庇护所过夜者（RR 3.44，95% CI 2.10-4.77），而居住在经济型旅馆者的风险也升高（RR 5.18，95% CI 1.14-9.23）。36种死因中33种（92%）的汇总RR估计值升高。死亡风险最高的是精神活性物质使用障碍（RR 21.38，95% CI 14.44-31.67）、意外伤害（RR 13.15，95% CI 5.46-31.69）和药物过量（RR 11.65，95% CI 6.97-19.48）。约一半的研究（57%）为高质量。未发现来自低收入国家的研究。没有发现发表偏倚的证据。 结论：与普通人群相比，无家可归者面临更高的过早死亡风险。最极端的不平等来自因精神活性物质使用障碍、事故和药物过量导致的死亡。针对精神活性物质使用和药物过量有循证治疗，表明其中一些死亡是可以避免的。 我们试图确定无家可归是否与全因和特定原因死亡风险相关。 我们发现，与未暴露于无家可归的人相比，暴露于任何形式无家可归的人全因死亡风险升高，且更可能死于精神活性物质使用障碍、药物过量、意外伤害和精神疾病。 无家可归以及其中许多死因是完全可避免的。 引言 在全球范围内，估计有近16亿人住房不足，每年约有1500万人被强制驱逐[]。在中低收入国家，城市化加剧以及冲突和气候变化导致的被迫流离失所，正在使没有永久住所的人数不断增加[]。在高收入国家，社会住房不足和私人住房日益难以负担，正导致无家可归者激增。据估计，欧洲每晚有895 000人——每600人中就有1人——露宿在收容所或临时住宿中，自2023–4年以来增加了40%[]。来自美国的相应数据表明，目前有771 480人正在经历无家可归，这比2023–4年增加了18%，其中大多数人居住在紧急收容所或临时住宿中，例如单人低价旅馆[]。 虽然现有研究提供了重要的证据基础[, ]，但其中的空白阻碍了它们对政策制定者的实用性，因为风险最高的无家可归经历类型、人口亚群和死因尚不清楚。首先，在以往唯一的一项荟萃分析中，作者将三项针对无家可归者的研究结果与针对曾被监禁者、性工作者或有物质使用障碍者的研究结果合并，因此未报告无家可归与全因死亡率之间的关联估计值[]。其次，关于风险是否因个体特征（如出生时性别）、无家可归经历类型（如露宿街头、收容所、廉价旅馆）或居住国家而异，目前知之甚少。第三，单项研究总结了少数死因的关联[, ]，给出了单一国家的死因别估计值（这可能限制普遍性）[, ]，或集中于高收入地区[]或某一种无家可归经历类型[]。最后，发表偏倚——即报告阳性结果的研究更有可能被发表这一观念——尚未被探讨，并且可能扭曲证据基础。 本研究旨在通过对无家可归与全因及死因别死亡率的全球证据进行系统回顾来弥补这些空白。我们选择的主要估计值是最多经调整的估计值以及使用社会经济剥夺对照组人群的估计值，因为这些可能最接近无家可归对死亡率影响的真实估计。 方法 本综述在PROSPERO（CRD42023430984）[]前瞻性注册，并按照已发表的方案[]进行。我们在报告呈现中遵循了PRISMA-IPD []和MOOSE []报告指南。 检索策略和纳入标准 我们通过系统检索MEDLINE、Embase、PsycINFO和Scopus，自建库至2024年12月确定了已发表的研究。完整检索词在（）中提供。未施加语言限制，如需，文章会被翻译成英文。 随后，我们获取了根据标题和摘要似乎符合纳入标准，或无法根据标题和摘要做出决定的研究的全文。我们还对系统综述的参考文献部分和网站进行了补充检索，以识别灰色文献。 在排除重复研究后，M.K.和A.W.独立审阅了所有标题和摘要，以根据既定的纳入标准 [] 确定其是否符合纳入条件。由于无家可归的定义在国际上有所不同 [, ]，我们采用了一个包容性的定义，即那些目前无家可归或报告曾经无家可归的人。这包括具有不同类型无家可归经历的人群：(i) 露宿街头（例如直接生活在街道、擅自占住的房屋中）；(ii) 住在为无家可归者提供的紧急旅社、收容所或避难所中；(iii) 居住在经济型酒店（例如包括单人房、单间出租房、寄宿旅馆在内的临时住所）；或 (iv) “沙发（couch）冲浪”（例如借住在朋友或亲戚家；此处称为沙发冲浪）。我们排除了以下研究：因经济无保障而面临（或曾经面临）失去住所风险的研究；居住条件较差的研究；或作者未提供足够信息以计算估计值的研究，例如那些仅量化无家可归人群死亡率而无对照组的研究。 任何分歧均通过讨论解决，必要时由第三位审稿人参与。当我们发现在不同出版物中呈现重复数据时，我们选择样本量最大或最具代表性的研究；当这些研究也相似时，我们使用最近发表的研究报告。 数据提取 所有作者提取了数据，随后对数据进行了复核。提取的特征包括：第一作者姓名、发表年份、研究地点（国家）、研究类型、无家可归类型、当前或既往无家可归状态、比较人群、社会经济劣势程度（如职业社会阶层、收入）、随访时间、暴露组和非暴露组的风险人数及死亡人数、报告的年龄和性别汇总统计量（如报告为频数、百分比，或均数或中位数及标准差或四分位数间距），以及统计调整的程度。 未发表的个体参与者数据 我们利用来自前瞻性队列研究的未发表个体层面数据，对已发表研究的数据进行了补充；这些队列研究是通过检索两个网站确定的，我们此前曾在这些网站上找到关于无家可归和死亡的数据 []，即英国数据服务（UK Data Service）(https://ukdataservice.ac.uk) 和大学间政治与社会研究联盟（Inter-University Consortium for Political and Social Research）(http://www.icpsr.umich.edu/icpsrweb/ICPSR/)。由此获得两项研究：欧洲健康、老龄化和退休调查 [] 与美国健康与退休研究 []。两项研究均获得伦理委员会批准，所有参与者均给予了知情同意。在我们对未发表研究的分析中，统一后的协变量包括年龄、性别和社会经济状况。 结局 结局为根据《国际疾病分类》第十版（ICD-10）定义的全因死亡率和死因别死亡率。 质量评估 为了评估纳入研究的质量，我们使用了流行病学研究的方法学标准（MASTER）量表[]。评估内容包括：正式招募、同等保留、同等确定、同等实施、同等预后、充分分析和时间先后性[最低得分=0，最高（高质量）=36]。如果总分高于中位数12，则研究质量被认为具有低偏倚风险。 统计分析 统计分析计划在分析前完成（补充文件，第6页）。所有统计分析的数据和代码可在GitHub上找到（https://github.com/jameswhite1979/Homelessness-systematic-review）。这些数据描述了每项研究所调查的死因。我们首先使用Cox比例风险回归分析未发表的个体参与者数据，以生成风险比（HRs）及其95%置信区间（CIs），用于评估无家可归与全因及特定原因死亡率之间的关联[]。 接下来，我们使用荟萃分析将未发表数据的分析结果与已发表研究的估计值合并。对于提供了风险比（HRs）和比值比（ORs）的已发表研究，我们将这些值用作相对危险度（RR）的合并估计值。在死亡率发生率<10%的研究中，我们将ORs视为RR的近似值[]。如果死亡率≥10%且有计数数据可用，则将ORs和HRs转换为RRs [, ]。如果置信区间（CIs）缺失，则使用标准公式计算。 我们分别分析了无家可归与全因死亡率和特定原因死亡率的关联。对于未发表的数据，效应估计值根据年龄、性别和社会经济状况进行了调整。对于已发表的研究，我们使用了调整最全面的估计值。一些出版物报告了多个效应量，造成了分析单位问题。为处理这一问题，当报告了多个比较人群组时，我们使用了以社会贫困人群为比较组，或对地区、家庭或个人层面贫困状况进行过调整的估计值。如果研究报告了多种比较（例如短期和长期随访），我们选择死亡人数最多的比较。当多个暴露与同一比较组进行比较时（例如不同类型的无家可归与同一未暴露人群），我们将其合并。当暴露组与不止一个比较组进行比较时（例如其他队列参与者或外部人群），我们进行了多水平元回归，对研究内的嵌套效应进行建模，以考虑共同组之间的关联[]。 我们使用随机效应模型，以反映纳入研究中预期的异质性。我们使用I2统计量来概括研究估计值中因异质性所致的总变异比例，数值越大表示异质性越大[]。计划中的探索性亚组分析按参与者性别、国家、无家可归状态（当前 vs. 既往）、无家可归者类型（露宿 vs. 收容所 vs. 临时住所）以及根据世界银行2023年分类[]所划分的研究样本来源国收入水平进行。我们还根据研究特征考察了差异，包括暴露确定方式（自我报告 vs. 行政记录）、对照人群（社会经济贫困人群 vs. 一般人群；内部队列 vs. 外部队列）、效应量调整（最小调整 vs. 多变量调整）、发表状态（未发表 vs. 已发表）、研究设计（回顾性 vs. 前瞻性）、总样本量（<1000，1000至<10 000，10 000至<100 000，100 000至<500 000，500 000至<100万，100万至<500万，500万至<1000万，≥1000万）以及研究质量（分类为低、高）。亚组差异通过使用元回归进行估计。 死亡原因的汇总相对风险（RRs）首先在ICD章节水平（如癌症、外部原因）进行估计，然后根据作者提供的描述词在更具体的死因分类层级（如肺癌、自杀）进行估计。通过漏斗图不对称性分析来探讨发表偏倚[]。在检测到发表偏倚时，我们使用非参数剪补法来量化其影响[]。由于我们预期会检索到少数规模非常大的研究，因此我们使用“留一法”[]来检验剔除每一项研究对汇总效应量的影响。 我们使用Stata（MP版本18）分析未发表数据，并使用Stata（MP版本18）和R（版本4.3.3）进行Meta分析。 结果 描绘了研究筛选过程。我们从四个电子数据库检索中识别出7630条记录，其中3274条重复。在并行补充检索符合条件的未发表数据集和灰色文献时，我们仔细检查了8项系统评价的参考文献部分、12个网站和16个在线数据库。这样做，额外识别出36条记录。我们审查了350项研究的全文，其中116项（114项已发表加上2项未发表数据集，如 所述）符合我们的纳入标准，涉及≥1.319亿人。被排除的研究及排除原因（附研究数量）列于 中。 研究筛选。 总共 1 618 049 名无家可归者和来自对照人群的估计 109 274 222 人参与了全因死亡率的荟萃分析。在 95 项研究中，全因死亡率分析得出了 126 个独立的相对危险度（RR），其中绝大多数（91.3%，N = 115）显示无家可归组的死亡风险升高。无家可归与全因死亡风险翻倍相关（RR 2.21，95% CI 1.91–2.57，P < .001；I2 = 99.7%；P < .001，）。略超过一半的研究（57.1%）被评价为高质量。我们没有发现发表偏倚的强证据（Egger 检验值 = 1.86，P = .06；）。 男性（RR 4.04，95% CI 2.87–5.21）和女性（RR 3.58，95% CI 2.35–4.82）之间，或者他们的经历是当前经历（RR 2.43，95% CI 1.92–2.95）与历史经历（RR 1.51，95% CI 0.80–2.22）相比，无家可归与死亡率之间的关联没有显著差异。在露宿者（RR 7.63，95% CI 3.29–11.97）中，这种关联比使用收容所的人（3.44，95% CI 2.10–4.77，P for difference = .01）更明显，且居住在廉价旅馆也与死亡风险升高相关（RR 5.18，95% CI 1.14–9.23），但由于只有一项研究报告了借宿沙发[]，我们在此未报告该结果。效应估计值也存在地域差异，来自欧洲（包括英国）的样本相对于美国（P for difference = .02）以及加拿大与美国相比（P = .003，）的 RR 更高。 无家可归与全因死亡率之间的关联，按人口亚组比较。 我们检查了各项研究的方法学特征，未发现由所采用的比较人群类型、暴露确定方法、模型中是否调整年龄或研究设计引起的偏倚的有力证据。使用外部比较人群的研究与使用内部比较人群的研究相比，汇总相对风险更高；提供调整后估计值的研究与提供未调整估计值的研究相比，汇总相对风险更高；低质量研究中的汇总相对风险更高；样本量较大的研究中的汇总相对风险也更高（样本量增加的趋势P < .001）；已发表研究的效应大于未发表研究 ()。 无家可归与全因死亡率之间的关联，按研究特征比较。 在所调查的36种死亡原因中，有33种（92%）的病因特异性死亡率的汇总估计值升高（无家可归者n = 1 202 205；比较者n = 25 089 695，）。在ICD-10章节层面上，精神和行为障碍的汇总相对风险特别高（11.78，95% CI 6.89–20.14，ICD-10中F00–F99）；损伤、中毒及其他外因后果（6.31，95% CI 4.34–9.18，ICD-10中S00–T98）；其他/不明原因死亡（5.56，95% CI 3.14–9.86，ICD-10中R00–R99）；血液及造血器官疾病（5.37，95% CI 1.19–24.17，ICD-10中D50–D89）；以及外因（5.91，95% CI 4.52–7.73，ICD-10中V01–Y98）。 无家可归与按ICD-10疾病类别划分的病因特异性死亡率之间的关联。 针对分类后各独立死因的汇总RR值中，最高的是归因于精神活性物质使用障碍的死亡（21.38，95% CI, 14.44–31.67）；药物过量（11.65，95% CI 6.97–19.48）；意外伤害的其他外部原因（13.15，95% CI 5.46–31.69）；意外事故（9.40，95% CI 5.38–16.42）；酒精相关原因（5.93，95% CI 1.60–22.04）；药物相关（不含药物过量）（4.37，95% CI 2.57–7.41）；他杀（6.31，95% CI 2.86–13.89）；以及自杀（4.11，95% CI 2.80–6.02）（）。 采用逐一剔除一项研究的敏感性分析方法来探究单项研究的影响，对大多数结果几乎没有影响。剔除一项在英国进行的研究[]（其标准化死亡比（SMR）是其他任何研究的2至3.7倍）后，欧洲的合并RR值从3.09（95% CI 2.21–3.97）降至2.39（95% CI 1.84, 3.11）。该研究的作者承认，无家可归者组的分母存在不确定性，因此SMR可能被高估。 讨论 我们的系统综述和荟萃分析发现，与一般对照人群相比，有无家可归史的人群死亡风险增加了一倍。男性和女性的风险幅度相当，无论是当前还是既往暴露；所有形式无家可归（包括露宿街头、居住于廉价旅馆或收容所）的风险均升高，而且来自加拿大的研究风险高于美国。我们发现，几乎所有被检查的健康状况死亡风险均升高，其中精神活性物质使用障碍的RR值极端升高（>20），药物过量和意外伤害外部原因RR为10–19，事故和酒精相关原因RR为5–10。 我们的研究结果与此前关于无家可归与全因死亡率的系统综述结果一致[, ]，那些综述的证据基础明显较小（11项研究），还有一项在单一高收入县开展的包含630万人的较大研究[]。在本综述中，我们纳入了116项研究，涉及至少1.319亿人。与此前一项针对无家可归类型的研究一致[]，我们发现所有类型（包括廉价旅馆）的死亡率均升高。这可能令政策制定者关注，因为近年来高收入国家在无法获得固定住所时，会长期使用廉价旅馆来安置单身成年人[]和寻求庇护者[, ]。重要的是，我们提供了全面的……
全文（开放获取） · 英文原文
【Abstract】 BACKGROUND: The association between homelessness and the type of homeless experience with all-cause and cause-specific mortality is uncertain. METHODS: Published studies were retrieved through a systematic search of MEDLINE, Embase PsycINFO, and Scopus from inception to December 2024. Unpublished data were identified from open-access data archives. We used random-effects meta-analysis to combine the effect estimates from published and unpublished data from 116 studies comprising 2 563 633 homeless people and 129 292 553 population controls. This review is registered at PROSPERO (CRD42023430984). RESULTS: The risk of all-cause mortality was increased in people exposed to homelessness compared with those unexposed [relative risk (RR) 2.21, 95% confidence interval (CI) 1.91-2.57, P < .001, I2 = 99.7], with risks similar in men (RR 4.04, 95% CI 2.87-5.21) and women (RR 3.58, 95% CI 2.35-4.82). Risks were higher in people who had slept rough (RR 7.63, 95% CI 3.29-11.97) than in those who had slept in homeless hostels/shelters (RR 3.44, 95% CI 2.10-4.77) and also elevated for people staying in low-cost hotels (RR 5.18, 95% CI 1.14-9.23). Summary RR estimates were elevated for 33 of the 36 (92%) causes of death. The highest risk of death was apparent for psychoactive-substance-use disorder (RR 21.38, 95% CI 14.44-31.67), accidental injuries (RR 13.15, 95% CI 5.46-31.69), and drug overdose (RR 11.65, 95% CI 6.97-19.48). Around half of the studies (57%) were of high quality. No studies were found in low-income countries. There was no evidence of publication bias. CONCLUSION: Compared with the general population, homeless people experience an increased risk of premature mortality. The most extreme inequalities were from deaths through psychoactive-substance-use disorder, accidents, and drug overdose. There are evidence-based treatments for psychoactive substance use and drug overdose, suggesting that some of these deaths are avoidable. We sought to determine whether homelessness was associated with the risk of all-cause and cause-specific mortality. We found that people exposed to all forms of homelessness had an elevated risk of all-cause mortality and were more likely to die from psychoactive-substance-use disorder, drug overdose, accidental injury, and mental illness compared with people who were not exposed to homeless. Homelessness and many of these causes of death are eminently avoidable. 【Introduction】 Worldwide, it is estimated that nearly 1.6 billion people have inadequate shelter, with ∼15 million people forcefully evicted every year []. In low- and middle-income countries, a combination of increasing urbanization and forced displacement from conflict and climate change is increasing the number of people without permanent shelter []. In high-income countries, insufficient social housing and increasingly unaffordable private housing are leading to a surge in homelessness. An estimated 895 000 people in Europe—one in every 600 people—sleep rough in shelters or temporary accommodation every night, which is an increase of 40% since 2023–4 []. Corresponding data from the USA suggests that 771 480 people are currently experiencing homelessness, which is an increase of 18% on 2023–4, with the majority living in emergency shelters or temporary accommodation such as single-occupancy low-cost hotels []. While existing studies have provided an important evidence base [, ], gaps hamper their usefulness for policy makers, as the types of homeless experience, population subgroups, and causes of death that are the most elevated are unclear. First, in the only previous meta-analysis, the authors combined the results from three studies with people who had been homeless with people who had been incarcerated, were sex workers, or had a substance-use disorder, such that no estimate of the association between homelessness and all-cause mortality was reported []. Second, little is known about whether the risk varies according to the characteristics of the individuals (e.g. sex at birth), type of homeless experience (e.g. rough sleeping, shelters, low-cost hotels), or country of residence. Third, individual studies have summarized associations for a small number of causes of death [, ], presented cause-specific estimates for a single country (which may limit generalizability) [, ], or focussed on high-income regions [] or one type of homeless experience []. Lastly, publication bias—the notion that studies reporting positive results are more likely to be published—has not been explored and can distort the evidence base. This study aimed to address these gaps by conducting a comprehensive review of the global evidence on homelessness with all-cause and cause-specific mortality. The primary estimates that we chose are those most adjusted and those using socio-economically deprived comparator groups, as these are likely to be the best estimates of the effect of homelessness on mortality. 【Methods】 The review was prospectively registered with PROSPERO (CRD42023430984) [] and conducted in accordance with a published protocol []. We followed the PRISMA-IPD [] and MOOSE [] reporting guidelines in the presentation of our report. 【Search strategy and selection criteria】 We identified published studies through a systematic search of MEDLINE, Embase, PsycINFO, and Scopus from their inception to December 2024. Full search terms are provided in the (). No limits on language were applied and articles were translated into English if required. Full texts were then retrieved of papers that appeared to meet the inclusion criteria or, where a decision could not be made, based on titles and abstracts. We also conducted a supplementary search of the reference sections of systematic reviews and websites to identify grey literature. After the exclusion of duplicate studies, M.K. and A.W. independently reviewed all titles and abstracts to establish their eligibility on the basis of defined inclusion criteria []. As definitions of homelessness vary internationally [, ], we used an inclusive definition of people who are, or report ever having been, homeless. This included people with different types of homeless experience: (i) sleeping rough (e.g. living directly on the streets, in squats); (ii) staying in an emergency hostel, shelter, or refuge for homeless people; (iii) residing in low-cost hotels (e.g. temporary accommodation including single-occupancy rooms, bedsits, boarding houses); or (iv) ‘sofa (couch)-surfing’ (e.g. staying with friend(s) or relative(s); herein called sofa-surfing). We excluded studies in which a residence was (or had been) threatened due to financial insecurity; if the people resided in poor living standards; or if the authors provided insufficient information to calculate an estimate, such as those studies solely quantifying mortality in a homeless population without a comparator group. Any disagreements were resolved through discussion, involving a third reviewer where necessary. Where we identified duplicate data presented in separate publications, we selected the study with the largest or most representative sample size and, when these were also similar, we used the most recently published report. 【Data extraction】 All authors extracted the data, which was then double-checked. Extracted characteristics included: name of the first author, publication year, study location (country), study type, homelessness type, current or historical homelessness, comparator population, degree of socio-economic disadvantage (e.g. occupational social class, income), follow-up period, number of people at risk and number of deaths in exposed and non-exposed groups, the summary statistics reported for age and sex (e.g. either reported as frequency, percentage, or mean or median alongside standard deviation or interquartile range), and the extent of statistical adjustment. 【Unpublished individual-participant data】 We supplemented data from the published studies with unpublished individual-level data from prospective cohort studies, located by searching two websites on which we had previously found data on homelessness and mortality [], the UK Data Service (https://ukdataservice.ac.uk), and the Inter-University Consortium for Political and Social Research (http://www.icpsr.umich.edu/icpsrweb/ICPSR/). This yielded two studies: the Survey of Health, Ageing and Retirement in Europe [] and the US Health & Retirement Study []. Both had approvals from ethics committees and all participants had given informed consent. The harmonized covariates in our analyses of unpublished studies included age, sex, and socio-economic status. 【Outcomes】 Outcomes were all-cause and cause-specific mortality defined according to the International Classification of Diseases, tenth revision (ICD-10). 【Quality assessment】 To assess the quality of the included studies, we used the MethodologicAl STandards for Epidemiological Research (MASTER) scale []. Assessments were made of: formal recruitment, equal retention, equal ascertainment, equal implementation, equal prognosis, sufficient analysis, and temporal precedence [minimum score = 0, maximum (high quality) = 36]. The quality of the study was regarded as having a low risk of bias if the total score was above a median of 12. 【Statistical analysis】 The statistical analysis plan was completed before analysis (Supplementary file, page 6). The data and code for all the statistical analyses can be found at GitHub (https://github.com/jameswhite1979/Homelessness-systematic-review). The data describe the causes of death investigated by each study. We first analysed unpublished individual-participant data by using Cox proportional-hazards regression to generate hazard ratios (HRs) with accompanying 95% confidence intervals (CIs) for the associations between homelessness and all-cause and cause-specific mortality []. Next, we used meta-analysis to combine the results from the analyses of the unpublished data with estimates from published studies. Where published studies provided HRs and odds ratios (ORs), these were used as a common estimate of relative risk (RR). In studies in which the incidence of mortality was <10%, we regarded the ORs as a close approximations of RR []. If the mortality incidence was ≥10% and count data were available, then ORs and HRs were converted into RRs [, ]. If CIs were missing, then they were calculated by using standard formulae. We analysed the associations of homelessness with all-cause and cause-specific mortality separately. For the unpublished data, effect estimates were adjusted for age and sex, and socio-economic status. For published studies, we used the most comprehensively adjusted estimates. Some publications reported multiple effect sizes, creating a unit-of-analysis issue. To account for this, when multiple comparator population groups were reported, we used estimates where the comparison group was a socially deprived population or where measures were adjusted for area, household, or individual-level deprivation. If studies reported multiple comparisons (e.g. short and longer follow-ups), we chose the comparison with the greatest number of deaths. Where multiple exposures were compared with the same comparator group (e.g. different types of homelessness to the same unexposed population), they were combined. Where the exposed group was compared with more than one comparator group (e.g. other cohort participants or an external population), we ran a multilevel meta-regression to model the effects nested within studies to account for the association between the common groups []. We used random-effects models reflecting the anticipated heterogeneity in the included studies. We used the I2 statistic to summarize the proportion of total variation in study estimates due to heterogeneity, with higher values denoting greater heterogeneity []. Planned exploratory subgroup analyses were conducted according to participant sex, country, homelessness status (current vs. historical), type of homeless (sleeping rough vs. shelter vs. temporary accommodation), and according to the income of the country from which the study sample was drawn based on the World Bank 2023 classification []. We also examined differences according to the study characteristics, including exposure ascertainment (self-report vs. administrative record), comparator population (socio-economically deprived vs. general; internal vs. external cohort), effect sizes adjustment (minimal vs. multivariable), publication status (unpublished vs. published), study design (retrospective vs. prospective), total sample size (<1000, 1000 to <10 000, 10 000 to <100 000, 100 000 to <500 000, 500 000 to <1 million, 1 million to <5 million, 5 million to <10 million, ≥10 million), and study quality (categorized as low, high). Subgroup differences were estimated by using meta-regression. Summary RRs for causes of death were first estimated at the ICD chapter level (e.g. cancer, external) and then at higher specific-cause resolution by using the descriptors provided by the authors (e.g. lung cancer, suicide). Publication bias was explored by using an analysis of asymmetry in funnel plots []. Where detected, we used the non-parametric trim-and-fill method to quantify its effect []. As we anticipated retrieving a small number of very large studies, we used the ‘leave-one-out’ method [] to examine the impact of excluding each on the summary effect size. We performed statistical analyses by using Stata (MP version 18) to analyse the unpublished data and Stata (MP version 18) and R (version 4.3.3) to compute the meta-analysis. 【Results】 depicts the process of study selection. We identified 7630 records from four electronic database searches, of which 3274 were duplicated. In a parallel supplementary search for eligible unpublished datasets and grey literature, we scrutinized the reference sections of 8 systematic reviews, 12 websites, and 16 online databases. In so doing, 36 additional records were identified. We reviewed the full texts of 350 studies, of which 116 (114 published plus 2 unpublished datasets, as described in ) met our inclusion criteria, comprising ≥131.9 million people. Excluded studies and reasons for exclusion with study counts are listed in . Study selection. A total of 1 618 049 people who were homeless and an estimated 109 274 222 from comparator populations contributed to the meta-analysis of all-cause mortality. Ninety-five studies yielded 126 unique RRs in the analyses of all-cause mortality, of which the large majority (91.3%, N = 115) had an elevated risk of death in the homeless group. Homelessness was associated with a doubling in the risk of all-cause mortality (RR 2.21, 95% CI 1.91–2.57, P < .001; I2 = 99.7%; P < .001, ). Just over half of the studies (57.1%) were evaluated as being of high quality. We found no strong evidence of publication bias (Egger’s test value = 1.86, P = .06; ). There were no marked differences in the homelessness–mortality association between men (RR 4.04, 95% CI 2.87–5.21) and women (RR 3.58, 95% CI 2.35–4.82) or whether their experience was current (RR 2.43, 95% CI 1.92–2.95) compared with historical (RR 1.51, 95% CI 0.80–2.22). A stronger association was apparent in people who had slept rough (RR 7.63, 95% CI 3.29–11.97) compared with in those using a shelter (3.44, 95% CI 2.10–4.77, P for difference = .01) and residence in a low-cost hotel was also associated with an elevated mortality risk (RR 5.18, 95% CI 1.14–9.23) but, as only one study was found to have reported on sofa-surfing [], we did not report on it here. There were also differences in effect estimates by geographical location, with higher RRs in samples drawn from Europe (including the UK) relative to the USA (P for difference = .02) and Canada compared with the USA (P = .003, ). Association between homelessness and all-cause mortality comparing population subgroups. We examined the methodological characteristics of the studies finding no strong evidence of bias arising from the type of comparator population used; the method of exposure ascertainment; whether age was adjusted for in the model or not; or the study design. Summary RRs were higher in studies that used an external compared with an internal comparator population; in studies that provided adjusted compared with unadjusted estimates; in lower-quality studies; and in studies with larger sample sizes (P for trend for increasing sample size < .001); published studies had larger effects than unpublished studies (). Association between homelessness and all-cause mortality comparing study characteristics. Summary estimates for cause-specific mortality were elevated for 33 of the 36 (92%) causes of death investigated (n homeless = 1 202 205; n comparator = 25 089 695, ). At the ICD-10 chapter level, summary RRs were particularly high for mental and behavioural disorders (11.78, 95% CI 6.89–20.14, F00–F99 in ICD-10); injury, poisoning, and other consequences of external causes (6.31, 95% CI 4.34–9.18, S00–T98 in ICD-10); other/unknown deaths (5.56, 95% CI 3.14–9.86, R00–R99 in ICD-10); diseases of the blood and blood-forming organs (5.37, 95% CI 1.19–24.17, D50–D89 in ICD-10); and external causes (5.91, 95% CI 4.52–7.73, V01–Y98 in ICD-10). Association between homelessness and cause-specific mortality according to the ICD-10 disease categories. Summary RRs for disaggregated individual causes were the highest for deaths attributed to psychoactive-substance-use disorder (21.38, 95% CI, 14.44–31.67); drug overdose (11.65, 95% CI 6.97–19.48); other external causes of accidental injury (13.15, 95% CI 5.46–31.69); accidents (9.40, 95% CI 5.38–16.42); alcohol-related causes (5.93, 95% CI 1.60–22.04); drug-related (excluding overdose) (4.37, 95% CI 2.57–7.41); homicide (6.31, 95% CI 2.86–13.89); and suicide (4.11, 95% CI 2.80–6.02) (). Sensitivity analyses using the leaving-one-study-out approach to explore the effect of individual studies had little impact on the majority of findings. Excluding one study conducted in the UK [], with Standardised Mortality Ratios (SMRs) 2–3.7 times larger than those of any other study, reduced the pooled RR for Europe from 3.09 (95% CI 2.21–3.97) to 2.39 (95% CI 1.84, 3.11). The authors of this study acknowledged uncertainty in the denominator for the homeless group such that the SMRs may have been inflated. 【Discussion】 Our systematic review and meta-analysis found a doubling in the risk of death in people with a history of homelessness relative to a general comparator population. The magnitude was comparable for men and women, by current or historical exposure, and elevated across all forms of homelessness, including rough sleeping and residence in a low-cost hotel or shelter, and higher in studies drawn from Canada compared with the USA. We found that the mortality was elevated for almost all health conditions examined, with some extreme RRs of >20 for psychoactive-substance-use disorder, 10–19 for drug overdose and external causes of accidental injury, and 5–10 for accidents and alcohol-related causes. Our findings are consistent with those of the previous systematic reviews of homelessness and all-cause mortality [, ] that included a substantially smaller evidence base (11 studies) as well as a comparatively large study with 6.3 million people conducted in a single high-income county []. In the present review, we included 116 studies comprising at least 131.9 million people. Consistently with the one previous study on the type of homelessness [], we found that mortality was increased across all types, including low-cost hotels. This is likely to be of interest to policy makers, as low-cost hotels have recently been used in high-income countries to house single adults [] and asylum seekers [, ] over long periods when settled accommodation is not available. Importantly, we provided a compreh