Kirkbride et al¹ provide a comprehensive, rigorous and thoughtful overview of the literature on social determinants of mental health, focusing on evidence for both causal effects of determinants and effectiveness of interventions. They also put forward a series of recommendations, focusing on how to prioritize prevention and intervention that attends to social justice and poverty alleviation with more rigorous study designs and greater data surveillance. I would like to add one additional recommendation: the need for community risk assessments.

The literature mostly reports risk ratios and beta values in cohort studies and intervention trials, from which conclusions can be drawn about relative magnitudes of risk elevation and reduction. Yet these measures largely do not provide information on which causes are most important, or which interventions are likely to be most effective, at the community level. Community risk assessments provide context- and community-specific information to set priorities, to be realistic about which interventions are worth the effort to implement, and to identify gaps where new, or more effective and scalable, interventions need to be developed.

The quantification of community risk has a long history^2-5, yet its application remains limited, especially in psychiatric epidemiology. What makes community risk assessments useful is that they combine prevalence and effect size, and make transparent assumptions that are often unacknowledged in more standard relative measures.

Across communities with low or high prevalence of exposure, the risk ratio or beta from a regression model may be similar: for instance, those who are exposed have twice the risk of the outcome compared to those unexposed. However, the public health impact of exposure may vary tremendously across the scenarios, and is captured in the population prevalence difference, the expected cases in the exposed, the population attributable fraction, and the number needed to harm. In a rare exposure scenario, for example, we could prevent just less than 10% of cases even if we perfectly implemented an intervention that removed 100% of the exposure. In a common exposure scenario, instead, almost half of cases could be prevented if we eliminated all the exposure.

Highlighting public health impact in our quantification of associations between exposure and outcome would push us as a field towards prioritizing intervention development for common exposures, including many of the social determinants of health reviewed by Kirkbride et al. Interventions with small relative effect sizes will have more impact for common exposures than interventions with large relative effect sizes for rare exposures. For example, Kirkbride et al cite a systematic review of the association between socioeconomic status and child/adolescent mental health⁶, which reported that the disorder prevalence difference between individuals with high and low socioeconomic status ranged from 8.9% and 13.2%, respectively, in the lower bound, to 15.9% and 33.4%, respectively, in the upper bound. While the risk ratios in lower and upper bound are similar, the public health implications are remarkably different. We can estimate that eliminating poverty would reduce ~10% of disorder in the lower bound scenarios, and ~20% – twice as many – in the upper bound. Considering interventions to reduce childhood/adolescent poverty, such as cash transfers, which have a meta-analyzed odds ratio of 0.72 for adolescent internalizing conditions⁷, we can then model the anticipated impact of their scaling in terms of potential public health impact, given the anticipated population attributable proportion.

Greater attention to assessment of community impact of exposures will also provide more rigor to our framing of interventions. Among Kirkbride et al's recommendations is to strengthen causal inference in research on social determinants of health. Commonly used estimates often make heroic counterfactual/potential outcome assumptions that cloud interpretation. For example, a risk ratio from a cohort study comparing depression incidence among those who are in poverty and those who are not compares two scenarios: the potential depression outcome if the entire population were in poverty versus the potential depression outcome if poverty were eliminated. While valuable for etiologic identification and elaboration, such measures are not directly relevant for public health, because an intervention to eliminate all poverty is unfortunately more of a thought experiment than a reality. Instead, targeted intervention effects that incorporate community risk⁸ allow one to estimate the proportion of the outcome that could be prevented if we were able to reduce exposure by a specific amount. We might frame a question around estimating what proportion of depression cases could feasibly be prevented if poverty were shifted by the estimated amount.

This type of exercise, landscaping the total community impact of exposure, and addressing the potential impact of interventions within that landscape, can bring together epidemiology and implementation science. As Kirkbride et al (and others⁹) note, the scalability of interventions that we know to work is a major barrier to improving community mental health. Assessment of community impact can provide additional analytic scaffolding to such statements.

A common critique to community risk assessments is that they are context-specific and will change based on location and time, but that is precisely the point. In the real world, we assess risk and implement interventions within specific contexts. Community risk assessments also clarify that our understanding of effect sizes and intervention effects, including relative measures, are anchored in time; what would be a high impact at one time point may not be as high at another time point, as prevalence and risk factors shift.

We know that the generalizability of exposure and intervention effects across time and place is an important component of public health science, yet discussions of these concepts are often relegated to a few sentences at the end of our papers. Community risk assessments provide a way to integrate knowledge about the specifics of contexts into our science. There may be contexts across the world where we do not have sufficient information about exposure or outcome prevalence to reliably estimate community risk; in those circumstances, it is prudent to conduct foundational epidemiological surveillance before attempting to implement specific interventions, to minimize unintended consequences and squandered resources.

While community risk assessments are of great value, like all measures they can also be misused and misunderstood. Further, risk factors are often synergistic, thus measures such as population attributable fraction sum well beyond 100%. Yet, the inclusion of community risk assessments would make synergy across risk factors more transparent; by only reporting relative measures of interaction (e.g., interaction betas, stratified risk and odds ratios), their synergistic impact on population mental health remains obscured.

In summary, Kirkbride et al provide a tremendous service to the field with their extensive and thorough review. It is incumbent on all of us, as scholars and researchers, to develop interventions that address social determinants of health at multiple levels, through a social justice approach. Such efforts will be aided by acknowledging differences in exposure and intervention effects across contexts, and quantitatively modeling them with the readily tools developed in epidemiology. We are only as effective as the numbers that we produce. In addition to rigorous study design, we should reliably assess community risk to maximally affect change.

Kirkbride 等人¹对心理健康社会决定因素的文献进行了全面、严谨和深思熟虑的概述，重点关注决定因素因果效应和干预措施有效性的证据。他们还提出了一系列建议，重点是如何通过更严格的研究设计和更大的数据监测，优先考虑预防和干预，关注社会正义和扶贫。我想补充一项额外建议：社区风险评估的必要性。

文献主要报道队列研究和干预试验中的风险比和贝塔值，从中可以得出风险升高和降低的相对幅度的结论。然而，这些措施在很大程度上没有提供关于哪些原因最重要，或者哪些干预措施可能在社区层面最有效的信息。社区风险评估提供特定背景和社区的信息，以设定优先事项，现实地了解哪些干预措施值得努力实施，并找出需要制定新的或更有效和可扩展的干预措施的差距。

社区风险的量化有着悠久的历史^2-5 ，但其应用仍然有限，特别是在精神流行病学方面。社区风险评估的有用之处在于，它们结合了患病率和效应大小，并做出了透明的假设，而这些假设在更标准的相对措施中往往未被承认。

在暴露率低或高的社区中，回归模型的风险比或贝塔值可能相似：例如，暴露者的结果风险是未暴露者的两倍。然而，暴露对公共健康的影响在不同情况下可能存在巨大差异，并且体现在人口患病率差异、暴露中的预期病例、人群归因分数以及造成伤害所需的数量中。例如，在罕见的暴露情况下，即使我们完美地实施了消除 100% 暴露的干预措施，我们也只能预防不到 10% 的病例。相反，在常见的暴露情况下，如果我们消除所有暴露，几乎可以预防一半的病例。

在我们对暴露与结果之间的关联进行量化时强调公共卫生影响，将推动我们作为一个领域优先考虑对常见暴露的干预措施开发，包括 Kirkbride 等人审查的许多健康的社会决定因素。相对效应量较小的干预措施对常见暴露的影响大于相对效应量较大的干预措施对罕见暴露的影响。例如，Kirkbride 等人引用了一项关于社会经济地位与儿童/青少年心理健康之间关系的系统综述⁶ ，该综述指出，在 2016 年，社会经济地位高和低的个体之间的疾病患病率差异分别为 8.9% 和 13.2%。下限为 15.9%，上限为 33.4%。虽然下限和上限的风险比相似，但对公共卫生的影响却截然不同。我们可以估计，在下限情景中，消除贫困将减少约 10% 的混乱，而在上限情景中，消除贫困将减少约 20%（两倍）。考虑到减少儿童/青少年贫困的干预措施，例如现金转移支付，其对青少年内化条件的荟萃分析优势比为 0.72 ⁷ ，然后我们可以根据潜在的公共卫生影响来模拟其规模化的预期影响，考虑到预期人口归因比例。

更多地关注评估暴露的社区影响也将使我们的干预措施框架更加严格。 Kirkbride 等人的建议之一是加强健康社会决定因素研究中的因果推理。常用的估计通常会做出夸张的反事实/潜在结果假设，从而影响解释。例如，一项比较贫困人群和非贫困人群抑郁症发病率的队列研究的风险比比较了两种情况：全体人口陷入贫困时的潜在抑郁结果与消除贫困时的潜在抑郁结果。虽然这些措施对于病因识别和阐述很有价值，但与公共卫生没有直接关系，因为不幸的是，消除所有贫困的干预措施更多的是一种思想实验而不是现实。相反，纳入社区风险⁸的有针对性的干预效果可以让我们估计，如果我们能够将接触风险减少到一定程度，则可以避免的结果比例。我们可能会提出一个问题，估计如果按估计的数量转移贫困，可以切实预防多少比例的抑郁症病例。

这种类型的活动可以美化暴露对整个社区的影响，并解决该环境中干预措施的潜在影响，可以将流行病学和实施科学结合起来。正如 Kirkbride 等人（和其他人⁹ ）指出的那样，我们知道有效的干预措施的可扩展性是改善社区心理健康的主要障碍。对社区影响的评估可以为此类陈述提供额外的分析框架。

对社区风险评估的一个常见批评是，它们是针对具体情况的，并且会根据地点和时间而变化，但这正是重点。在现实世界中，我们评估风险并在特定情况下实施干预措施。社区风险评估还阐明，我们对效应大小和干预效果（包括相关措施）的理解是及时锚定的；随着患病率和风险因素的变化，在一个时间点产生很大影响的因素在另一个时间点可能不会那么大。

我们知道，不同时间和地点的暴露和干预效果的普遍性是公共卫生科学的重要组成部分，但对这些概念的讨论往往被限制在我们论文末尾的几句话中。社区风险评估提供了一种将有关具体环境的知识整合到我们的科学中的方法。在世界各地，我们可能没有足够的关于暴露或结果流行率的信息来可靠地估计社区风险；在这种情况下，明智的做法是在尝试实施具体干预措施之前进行基础流行病学监测，以尽量减少意外后果和资源浪费。

虽然社区风险评估具有很大的价值，但与所有措施一样，它们也可能被滥用和误解。此外，风险因素通常是协同作用的，因此人口归因分数之和等指标远远超过 100%。然而，纳入社区风险评估将使跨风险因素的协同作用更加透明；通过仅报告相互作用的相对测量（例如，相互作用贝塔、分层风险和比值比），它们对人口心理健康的协同影响仍然模糊。

总之，Kirkbride 等人通过广泛而彻底的审查为该领域提供了巨大的服务。作为学者和研究人员，我们所有人都有责任通过社会正义方法制定干预措施，在多个层面解决健康问题的社会决定因素。承认不同背景下暴露和干预效果的差异，并使用流行病学中开发的现成工具对其进行定量建模，将有助于此类努力。我们的效率取决于我们生产的数字。除了严格的研究设计之外，我们还应该可靠地评估社区风险，以最大程度地影响变革。