Risks of mental health outcomes in people with covid-19: cohort studyBMJ 2022; 376 doi: https://doi.org/10.1136/bmj-2021-068993 (Published 16 February 2022) Cite this as: BMJ 2022;376:e068993
All rapid responses
Rapid responses are electronic comments to the editor. They enable our users to debate issues raised in articles published on bmj.com. A rapid response is first posted online. If you need the URL (web address) of an individual response, simply click on the response headline and copy the URL from the browser window. A proportion of responses will, after editing, be published online and in the print journal as letters, which are indexed in PubMed. Rapid responses are not indexed in PubMed and they are not journal articles. The BMJ reserves the right to remove responses which are being wilfully misrepresented as published articles or when it is brought to our attention that a response spreads misinformation.
From March 2022, the word limit for rapid responses will be 600 words not including references and author details. We will no longer post responses that exceed this limit.
The word limit for letters selected from posted responses remains 300 words.
Electronic health record data for identifiyng COVID-19 long-term sequelae must be interpreted cautiously
I read with interest the study by Xie et al. concerning the neuropsychiatric sequelae of COVID-19 survivors (1). In related publications (2), the authors reported an increased incidence of cardiovascular (3), metabolic (4), and renal (5) long-term conditions among COVID-19 survivors. While these reports are important, there are open questions regarding data analysis that cast serious doubts on the authors’ main conclusions.
It is of utmost importance to understand that EHR data do not only reflect the patients’ health, but their interactions with the health care system (6). For example, surveillance bias implies that patients who are followed-up with increased vigilance are more likely to receive EHR diagnoses (8,9). In other words, an EHR diagnosis must not only be interpreted as a proxy for sickness, because it might also be a proxy for diagnostic intensity. Closely related, informed presence bias implies that patients with higher frequency of healthcare encounters, receive more EHR diagnoses, due to repeated interactions with the healthcare system (10), which necessitates a code for these encounters. In fact, some researchers believe that the predictive value of underlying healthcare processes might even be more relevant than the predictive value of the pathophysiological processes (6).
I reassessed the data of this study and found evidence of different healthcare utilization between exposure and control group. The authors report that 50.95% of COVID-19 patients had three or more encounters with the healthcare system vs. 27.83% of the control group. Fortunately, the authors used “frequency of outpatient encounters” as a covariate and achieved balance after matching. Nonetheless, according to the raw data in Nature, COVID-19 patients received more significantly more EHR codes for “medical examinations”, “encounters for administrative purposes” and reception of “diagnostic agents”. The latter clearly implies larger diagnostic work-up of COVID-19 patients. Of course, this might indicate poorer health of COVID-19 patients in general, however to the absolute minimum, the possible role of these biases and their impact on the measured findings must be highlighted as a major limitation of all conclusions in this study.
The authors contradict themselves in their published reports. In the Nature study, substance use disorders are not associated with COVID-19, except a positive trend for alcohol-related disorders. In the BMJ study, they do. Further baseline characteristics of COVID-19 patients differed dramatically. For example, after matching, the BMJ study reports 23.95% of COVID 19 patients suffering from diabetes mellitus type 2, vs. 29.27% in Nature vs. 24.39% in the Nature Medicine or vs. 39.4% in the JASN report, respectively. To the authors defense, all studies differ in statistical adjustment and selection of covariates but the degree of model dependence is concerning.
Strangely enough, in the Nature study, the authors show that COVID-19 is inversely associated with tobacco- and cannabis-related disorders, which is evidently implausible and clearly indicates residual confounding. What is most concerning, however, is that in the BMJ report the authors report substance use disorders – except tobacco-related disorders, which is the only outcome indicating uncontrolled confounding. The authors provide no rationale why they leave out the data on nicotine adherence.
Lastly, the authors used traumatic events and neoplasms as negative-outcome controls. The authors state that “if there were biases in the analytical approach, this would extend to the chosen negative-outcome controls”. However, the authors only chose outcomes, which align with their hypothesis. According to the raw data in Nature, COVID 19 is also associated with “refractive errors” and “acquired foot deformities”. It remains unclear, why those are not considered negative-outcome controls, as they are clearly not “plausibly associated to COVID-19 infections”.
The conclusions of this author group is hampered by severe uncontrolled confounding and needs careful reevaluation. To optimally assess the epidemiological burden of long-term sequelae of COVID-19, we must not rely on self-report or observational data, but on gold-standard prospective, double-blinded cohort studies (14).
I declare no competing interests.
The opinions expressed here do not necessarily represent the opinions of my affiliations.
1. Xie, Y., Xu, E. & Al-Aly, Z. Risks of mental health outcomes in people with covid-19: cohort
study. BMJ e068993 (2022) doi:10.1136/bmj-2021-068993.
2. Al-Aly, Z., Xie, Y. & Bowe, B. High-dimensional characterization of post-acute sequelae of
COVID-19. Nature 594, 259–264 (2021).
3. Xie, Y., Xu, E., Bowe, B. & Al-Aly, Z. Long-term cardiovascular outcomes of COVID-19. Nature
Medicine 28, 583–590 (2022).
4. Xie, Y. & Al-Aly, Z. Risks and burdens of incident diabetes in long COVID: a cohort study. The
Lancet Diabetes & Endocrinology 10, 311–321 (2022).
5. Bowe, B., Xie, Y., Xu, E. & Al-Aly, Z. Kidney Outcomes in Long COVID. Journal of the American
Society of Nephrology 32, 2851–2862 (2021).
6. Agniel, D., Kohane, I. S. & Weber, G. M. Biases in electronic health record data due to
processes within the healthcare system: retrospective observational study. BMJ k1479 (2018)
7. Hripcsak, G., Albers, D. J. & Perotte, A. Parameterizing time in electronic health record studies.
Journal of the American Medical Informatics Association 22, 794–804 (2015).
8. Haut, E. R. Surveillance Bias in Outcomes Reporting. JAMA 305, 2462 (2011).
9. Pierce, C. A. et al. Surveillance Bias and Deep Vein Thrombosis in the National Trauma Data
Bank: The More We Look, The More We Find. Journal of Trauma: Injury, Infection & Critical
Care 64, 932–937 (2008).
10. Goldstein, B. A., Bhavsar, N. A., Phelan, M. & Pencina, M. J. Controlling for Informed Presence
Bias Due to the Number of Health Encounters in an Electronic Health Record. American
Journal of Epidemiology 184, 847–855 (2016).
11. Vai, B. et al. Mental disorders and risk of COVID-19-related mortality, hospitalisation, and
intensive care unit admission: a systematic review and meta-analysis. The Lancet Psychiatry 8,
12. Mena, G. E. et al. Socioeconomic status determines COVID-19 incidence and related mortality
in Santiago, Chile. Science (1979) 372, (2021).
13. Lipsitch, M., Tchetgen Tchetgen, E. & Cohen, T. Negative Controls. Epidemiology 21, 383–388
14. Sneller, M. C. et al. A Longitudinal Study of COVID-19 Sequelae and Immunity: Baseline
Findings. Annals of Internal Medicine 175, 969–979 (2022)
Competing interests: No competing interests
We read the article by Xie and colleagues with great interest , however, we warn against a causal interpretation of the findings. Our research shows that propensity to ‘get tested’ and other factors are major confounders, and means we still do not have the evidence to conclude that COVID-19 infection can be a cause of psychiatric illness, in the way it clearly is for fatigue.
Our national study of routinely collected, English primary care data found a strong association between receiving a positive PCR COVID-19 test and subsequent psychiatric morbidity, versus controls without a positive test (adjusted Hazard Ratio 1.83, 95% CI 1.66-2.02) . However, we also found similar increases among those who received a negative test (aHR 1.71, 1.65-1.77). Similarly, a study of Danish registry data found little evidence of elevated risk of subsequent psychiatric outcomes among those with a positive COVID-19 test, when compared to those who tested negative  (adjusted risk ratio depression 0.91, 0.46 to 1.80; aRR anxiety disorders 0.54, 0.32 to 0.90).
We believe there are two sources of residual confounding that might be implicated. The first results from the selection of those who present for testing. In the US, between February 2020 and September 2021, an estimated 25% of those who had COVID-19 presented to testing and the figure for symptomatic cases is 29% . The majority that do not attend for testing appear to be very different to those that do; and we reported that did not present for testing were less likely to have a prior mental illness (eTable 5 in the online supplement ). The second source of confounding is that the virus itself targets particular groups. For example, those in lower-socioeconomic areas and those who have worked with the public throughout the pandemic were risk factors for both infection transmission  and for poor mental health, particularly during the pandemic . Accounting for these two sources of confounding is challenging in registry data, where data on mental health and socio-economic context are often crudely measured or not available.
Taken together, therefore, we caution against making causal inferences about COVID 19 infection and subsequent mental health because there is good evidence for strong confounding in analyses of those who present for COVID-19 testing; and this is likely to bias studies of COVID-19 and psychiatric sequelae. In order to improve our understanding of whether COVID-19 infection directly increases the risk of mental illness, we recommend using methods or data sources that might be less subject to confounding bias. For example, using random population-based cohorts enrolled in seroprevalence studies. Further careful work is needed to understand the links between incident mental illness and people’s experiences during a pandemic, including trauma following hospitalisation or critical care and how this intersects with demographic risk factors for infection such as ethnicity and socioeconomic factors. This will be important as future pandemics with new pathogens are increasingly likely in our globalised world.
1 Xie Y, Xu E, Al-Aly Z. Risks of mental health outcomes in people with covid-19: cohort study. BMJ 2022;376:e068993. doi:10.1136/bmj-2021-068993
2 Abel KM, Carr MJ, Ashcroft DM, et al. Association of SARS-CoV-2 Infection With Psychological Distress, Psychotropic Prescribing, Fatigue, and Sleep Problems Among UK Primary Care Patients. JAMA Netw Open 2021;4:1–14. doi:10.1001/jamanetworkopen.2021.34803
3 Lund LC, Hallas J, Nielsen H, et al. Post-acute effects of SARS-CoV-2 infection in individuals not requiring hospital admission : a Danish population-based cohort study. 2021;3099:1–10. doi:10.1016/S1473-3099(21)00211-5
4 Centres for Disease Control and Prevention. Estimated COVID-19 Burden. CDC 24/7 Sav. Lives Pretecting People. 2021.https://www.cdc.gov/coronavirus/2019-ncov/cases-updates/burden.html (accessed 23 Aug 2021).
5 Larsen T, Bosworth M, Nafilyan V. Coronavirus (COVID-19) case rates by socio-demographic characteristics. ONS. London: 2021. https://www.ons.gov.uk/peoplepopulationandcommunity/healthandsocialcare/...
6 Pierce M, McManus S, Hope H, et al. Mental health responses to the COVID-19 pandemic: a latent class trajectory analysis using longitudinal UK data. The Lancet Psychiatry 2021;8:610–9. doi:10.1016/s2215-0366(21)00151-6
Competing interests: No competing interests
Because sequelae included preexisting conditions, I wonder if it is possible that dementia was already present before COVID-19 in most seniors in this study that had dementia after COVID-19. If so, because COVID-19 outbreaks were common in senior residences, could reverse causality be possible? Being an institutionalized senior more likely to have dementia increased risk of COVID?
Competing interests: No competing interests
This is very well designed research which was based on a large sample size, and it presents unique insights into the link between catching covid-19 and the risk of mental disorders. In interpreting these findings, general practitioners (GPs) and psychiatrists must be aware that the findings shed important light on the possibility that covid-19 infection is a risk factor for certain mental disorders among some people, but the relationship is likely to be indirect and not direct.
During diagnostic formulation when working with patients who present with mental disorder symptoms, clinicians should only consider covid-19 in the sense that it raises the risk of psychological variables which raise the risk of mental disorders, e.g., loneliness and access to social support. Covid-19 made many patients lonely  because they had to self-isolate as a mechanism of infection control. Being isolated likely reduced many patients' access to social support, and the infection gave many patients impaired breathing and debilitating fatigue , making exercise and regular activities difficult. Facing such drastic limits to their daily life, in turn, likely reduced patients’ feeling of self-efficacy, making them feel purposeless and unfulfilled with daily life. As well, patients with covid-19 were likely distressed about being infected because it was a stressful thing for many. 
The problems likely continued for a long time because many patients experienced symptoms that lasted a while , and the impairments to daily life might have exacerbated psychological harm and raised the risks of mental disorders reported in this study. The average age of participants in this study was 63 years and likely comprised many non-working participants which might have worsened their social isolation and reduced how much social support they had after catching covid-19 (because many working adults continued to work remotely). Although the researchers compared them with people of an equivalent age, those who did not have covid-19 were more likely to have continued with daily activities (e.g., social contact, going outside and exercising) beneficial to their mental health. The topics of social isolation and loneliness among older adults have inspired many studies, including interventions , therefore the age of the participants is an important consideration to clinicians interpreting the study findings. This means that the mental disorder risk ratios found in this study might not apply to the working-age population (for whom the levels of risk might be smaller), but the findings are nonetheless very useful in shedding light on this issue.
 Banerjee D, Rai M. Social isolation in Covid-19: The impact of loneliness. Int J Soc Psychiatry. 2020;66(6):525-527. doi:10.1177/0020764020922269
 Sudre, C.H., Murray, B., Varsavsky, T. et al. Attributes and predictors of long COVID. Nat Med. 2021; 27, 626–631.
 Daly M, Robinson E. Psychological distress and adaptation to the COVID-19 crisis in the United States. J Psychiatr Res. 2021; 136:603-609.
 Cattan M, White M, Bond J, Learmouth A. Preventing social isolation and loneliness among older people: A systematic review of health promotion interventions. Ageing Soc. 2005; 25:41-67.
Competing interests: No competing interests