This journal recently published a paper by Rong et al., entitled ‘Persistent moderate to severe pain and long-term cognitive decline’. (Rong et al., 2021). The authors demonstrate that, to a small but statistically significant degree, older adults with persistent moderate-to-severe pain (an approximation for chronic pain) experience a faster rate of late-life cognitive decline than older adults without pain. Given these findings, should clinicians be alert for accelerated cognitive decline in older adults with chronic pain? Rong et al. argue yes. But, interestingly, using the same data source, an earlier study by Veronese et al., (2018) concluded that there was no evidence for an effect. What are we to make of this?

Rong and Veronese both used the English Longitudinal Study of Ageing (ELSA), a large population-based study of English older adults, to measure whether there exists evidence for a faster rate of cognitive decline in participants who report moderate-to-severe pain. Veronese et al. conducted a short-term analysis (2 years of cognitive change) and concluded that pain was not associated with accelerated cognitive decline. In contrast, Rong et al. found a highly statistically significant association that was replicated in multiple subgroup/sensitivity analyses. Rather than being inherently contradictory, important differences in the analytic methods and the handling of non-response likely explain the differing conclusions of Rong and Veronese. Rong adopted an analytic strategy similar to that of Whitlock et al., (2017), my group, when we studied the impact of persistent pain on cognition in an aligned cohort of US older adults, reaching qualitatively identical conclusions on the primary outcome of interest. Thus, Rong et al. externally replicated and, concurrently, internally repudiated the existing manuscripts, elegantly demonstrating the self-correcting and iterative nature of science.

Whilethere were differences between Rong and Whitlock in elements of the study, including the way cognitive tests were handled, the adjustment covariates, andthe characterization of persistent pain, the overall analytic approach was thesame between the two studies: a linear mixed effects model with random effectsat the participant level, allowing for a different rate of change over time(i.e. slope) for those with persistent pain. This is an intuitive strategy which does not assume that everyone experiences the same underlying rate of cognitive decline or the same ‘acceleration’ (with apologies for causal language!) with chronic pain. Studying a long duration of cognitive test results—10 years in Whitlock et al, and 12 years in Rong et al—allows for more precise estimates, and may explain similarities between Rong and Whitlock's conclusions. Rong goes further, as well, to look for a dose–response relationship between pain severity and rate of decline (it does exist), to test whether there was a difference in young-old versus older-old individuals (there was not) and to test whether certain anatomic pain sites behave differently (they do not).

There is one other crucial difference between the approach of Rong versus that of Veronese: the handling of non-response. In any longitudinal study, some participants drop out—often because of death or because they no longer wish to participate. In an analysis of chronic pain and cognition, it must be assumed that that dropout is strongly associated with either the primary predictor (pain) and/or the primary outcome (cognition). That is, older adults who are constantly in pain, or who are cognitively unable to participate in testing, may be more likely to die, may elect to stop contributing cognition data, or may exit the study. This will, as Rong notes, most likely bias the findings to the null (i.e. create a bias that makes it more likely to conclude there was no association)—the most affected participants, who would strengthen the association between chronic pain and low cognition, drop out and cannot be observed. If bias towards the null is expected from dropout, any statement of ‘no difference’, as Veronese concluded, should be viewed with a particularly critical eye. If we magically corrected dropout noted in Rong's work, we would expect an even stronger association between baseline persistent pain and accelerated long-term cognitive decline.

Neither Rong nor Whitlock found a particularly devastating strength of association. If older adults with chronic pain experience accelerated cognitive decline—as seems likely, given the external replication of these findings—then the expected population-average impact on independence would be subtle. Rong estimates that it would take 17 years to reach clinical significance. Whitlock extrapolated that the difference in cognition after 10 years translated to an absolute increased risk of about 2% for being unable to conduct cognitively intensive instrumental activities of daily living. Nonetheless, as pain is a potentially modifiable factor, and given the importance of maintaining optimal cognitive health upon older adults’ independence as they age, further studies to better understand the causal nature of this reproducible association are unequivocally warranted.

该杂志最近发表了 Rong 等人的一篇论文，题为“持续中度至重度疼痛和长期认知衰退”。（荣等人，  2021 年）。作者证明，在较小但具有统计学意义的程度上，持续中度至重度疼痛（慢性疼痛的近似值）的老年人比没有疼痛的老年人经历的晚年认知下降速度更快。鉴于这些发现，临床医生是否应该警惕患有慢性疼痛的老年人的认知能力加速下降？荣等人。说是的。但是，有趣的是，使用相同的数据源，Veronese 等人（2018 年）的早期研究得出结论，没有证据表明有影响。我们该怎么做？

Rong 和 Veronese 都使用了英国老龄化纵向研究 (ELSA)，这是一项针对英国老年人的大型基于人群的研究，以衡量是否有证据表明报告中度至重度疼痛的参与者的认知能力下降速度更快。委罗内塞等人。进行了一项短期分析（2 年的认知变化）并得出结论，疼痛与认知能力加速下降无关。相比之下，Rong 等人。发现了在多个亚组/敏感性分析中重复的高度统计学显着的关联。分析方法和不回复处理的重要差异可能解释了荣和委罗内塞的不同结论，而不是本质上的矛盾。Rong 采用了类似于 Whitlock 等人的分析策略，（2017 年），我的小组，当我们研究持续性疼痛对美国老年人群的认知影响时，就感兴趣的主要结果得出了定性相同的结论。因此，Rong 等人。外部复制，同时内部否定现有手稿，优雅地展示了科学的自我纠正和迭代性质。

虽然 Rong 和 Whitlock 在研究的要素上存在差异，包括认知测试的处理方式、调整协变量和持续性疼痛的表征，但两项研究的总体分析方法是相同的：具有随机效应的线性混合效应模型参与者级别，允许持续性疼痛患者随时间变化的不同速率（即斜率）。这是一个直观的策略，它不假设每个人都经历相同的认知衰退率或相同的“加速”（为因果语言道歉！）慢性疼痛。研究长期的认知测试结果——Whitlock 等人 10 年，Rong 等人 12 年——可以进行更精确的估计，并可以解释 Rong 和 Whitlock 的结论之间的相似之处。

Rong 的方法与 Veronese 的方法之间还有另一个重要的区别：对不回复的处理。在任何纵向研究中，都会有一些参与者退出——通常是因为死亡或不再希望参与。在对慢性疼痛和认知的分析中，必须假设辍学与主要预测因子（疼痛）和/或主要结果（认知）密切相关。也就是说，长期处于疼痛状态或认知无法参与测试的老年人可能更有可能死亡，可能会选择停止提供认知数据，或者可能退出研究。正如 Rong 指出的那样，这很可能会使研究结果偏向于零（即产生一种偏见，使其更有可能得出没有关联的结论）——受影响最大的参与者，谁会加强慢性疼痛与认知能力低下之间的关联，辍学且无法观察。如果预期辍学会偏向零，那么正如 Veronese 总结的那样，任何“没有区别”的陈述都应该以特别批判的眼光看待。如果我们神奇地纠正了 Rong 工作中提到的辍学，我们预计基线持续性疼痛与加速的长期认知衰退之间的关联会更强。

荣和惠特洛克都没有发现一种特别具有破坏性的联想力量。如果患有慢性疼痛的老年人认知能力下降加速——考虑到这些发现的外部复制，这似乎很可能——那么预期的人口平均对独立性的影响将是微妙的。荣估计要达到临床意义需要 17 年。Whitlock 推断，10 年后认知的差异转化为无法进行日常生活中认知密集型工具性活动的绝对风险增加约 2%。尽管如此，由于疼痛是一个潜在的可改变因素，并且考虑到在老年人随着年龄的增长而独立时保持最佳认知健康的重要性，