Cardiovascular disease and subsequent risk of psychiatric disorders: a nationwide sibling-controlled study

Article activity feed

1. 

   Version published to 10.7554/elife.80143 on eLife

   Oct 21, 2022
2. 

   eLife

   Sep 26, 2022

   Author Response

   Reviewer #1 (Public Review):

   This is a well-done analysis using the very robust Swedish national population registry.

   The study strengths include large size, prolonged follow-up, and use of two comparison populations.

   Thank you for the encouraging comments on our study.

   The main limitations which need to be addressed by the authors are accounting for reverse causality, namely if a psychiatric illness (PI) developed before or about the same time as the CVD. The much steeper risk relationships early after a CVD event are so suggestive. Some further analyses to tease out those with clearly NO PI before CVD would be in order.

   Thank you for the comment. Previous studies have consistently reported an association between psychiatric disorders and CVD [1,2], thus, we agree that reverse causality may, in principle, …

   Author Response

   Reviewer #1 (Public Review):

   This is a well-done analysis using the very robust Swedish national population registry.

   The study strengths include large size, prolonged follow-up, and use of two comparison populations.

   Thank you for the encouraging comments on our study.

   The main limitations which need to be addressed by the authors are accounting for reverse causality, namely if a psychiatric illness (PI) developed before or about the same time as the CVD. The much steeper risk relationships early after a CVD event are so suggestive. Some further analyses to tease out those with clearly NO PI before CVD would be in order.

   Thank you for the comment. Previous studies have consistently reported an association between psychiatric disorders and CVD [1,2], thus, we agree that reverse causality may, in principle, explain some of the observed results indicating a rise in incident psychiatric disorders after incident CVD, particularly during the immediate period. Yet, it is reasonable to assume that a diagnosis of a lifethreatening disease, such as CVD, is in many cases a traumatic experience resulting in an immediate rise in risks of psychiatric disorders. Others have reported such associations e.g. after natural disasters and we have indeed observed such a pattern in our previous work, e.g., after cancer diagnosis [3]. However, we agree that reverse causality cannot be excluded and may partly contribute to the highly increased risk of psychiatric disorder immediately after CVD diagnosis. Indeed, some of these patients may have been attended for their psychiatric disorders in primary care before the incident CVD. As the Patient Register only captures in- and outpatient hospital care, we have conducted an additional analysis, also excluding individuals with previous prescriptions of psychotropic drugs (ATC codes: N05, N06) before their incident CVD – thereby adding a detection of patients with prevalent mental health problems attended by primary care. The results show similar point estimates (Supplementary Appendix Table S5, listed also as below) thus not supporting the notion that reverse causality is a major concern. Furthermore, the association is noted up to 28 years after CVD diagnosis, which is unlikely due to reverse causality.

   We have now added our motivation for this additional analysis on the Method (Page 9), as below. “Because the Swedish Patient Register includes only information related to specialist care, we might have misclassified patients with a history of milder psychiatric disorders diagnosed before index date attended only in primary care. To account for the reverse causality of having undetected psychiatric disorders or symptoms before the incident CVD, we performed a sensitivity analysis additionally excluding study participants with prescribed use of psychotropic drugs before the index date (ascertained from the Swedish Prescribed Drug Register including information on all prescribed medication use in Sweden since July 2005), and followed the remaining participants from 2006 to 2016.”

   Second, for the robust matched cohort design, the authors age and sex matched each patient with 10 individuals from the general population and then also stratified their model by the matching variables. Could adjusting for matched factors in such cohort studies re-introduce bias into these estimates?

   Thank you for the comment. Adjusting for matching factors should provide estimates with the same validity as using a stratified model. In our study, we matched individuals diagnosed with a CVD with their unaffected full siblings as well as 10 randomly selected, unexposed individuals, on the same age and sex, without such diagnosis. As controlling for matching variables is recommended when there are additional confounders [1,2], we used a stratified Cox model commonly applied in family-based studies [3,4].

   References:

   1.Sjölander A, Greenland S. Ignoring the matching variables in cohort studies - when is it valid and why? Stat Med. 2013 Nov 30;32(27):4696-708.
   2.Mansournia MA, Hernán MA, Greenland S. Matched designs and causal diagrams. Int J Epidemiol. 2013 Jun;42(3):860-9.
   3.D'Onofrio BM, Lahey BB, Turkheimer E, Lichtenstein P. Critical need for family-based, quasiexperimental designs in integrating genetic and social science research. Am J Public Health. 2013 Oct;103 Suppl 1(Suppl 1):S46-55.
   4.Song, H., Fang, F., Arnberg, F. K., Mataix-Cols, D., de la Cruz, L. F., Almqvist, C., ... & Valdimarsdóttir, U. A. (2019). Stress related disorders and risk of cardiovascular disease: population based, sibling controlled cohort study. bmj, 365.

   Third, the range of PIs associated with CVD is a lot broader than would be expected or unexpected (eg eating disorders!).

   Thank you for the comment. We agree with the reviewer that the strong association between CVD and incident eating disorders is somewhat surprising although the link between cardiovascular risk factors (e.g. obesity) and binge eating have indeed been reported [1,2]. We have now performed the analysis on the association between first-onset CVD and following incident eating disorder, additionally excluding individuals with a history of psychotropic medication use. We found that the associations became even stronger after this exclusion (Supplementary table 5). It is possible that individuals suffering their first CVD indeed drastically alter their lifestyle, in some cases resulting in dysfunctional eating and may therefore be vulnerable to eating disorders. Given that the evidence assessing the risk of eating disorder among CVD patients is still limited, our study adds a valuable piece of knowledge on this regard and calls for further investigations to better understand this association.

   References:

   1.Mitchell JE. Medical comorbidity and medical complications associated with binge-eating disorder. Int J Eat Disord. 2016 Mar;49(3):319-23.
   2.Bulik CM, Sullivan PF, Kendler KS. Medical and psychiatric morbidity in obese women with and without binge eating. Int J Eat Disord 2002;32:72–78.

   Lastly, the authors should try to account for secular changes in smoking and alcohol consumption or BMI over the study period. In particular, while Sweden never had very high smoking rates (due to Snus) alcohol use within specific cohorts might have both affected CVD risk (particularly stroke) and PI risk. Examining trends in for example liver cirrhosis over the study time period might help or use sales/consumption data. The authors do recognize a limitation in being unable to adjust for smoking, alcohol, and adiposity.

   Some additional analyses to address these points and some more caution in the discussion are required.

   Thank you for the comment. As the reviewer points out, we do recognize the potential unmeasured influence of lifestyle factors (e.g. smoking and alcohol consumption) on the studied associations as these data are not collected in the Swedish registries. However, the associations between CVD and psychiatric disorders were quite stable across calendar time, although somewhat stronger by the end of the observation period. The evidence does not suggest a drastic change in lifestyle factors in Sweden during the latter part of the observation period except for a slight increase in alcohol consumption [1,2] and liver cirrhosis [3]. Although we find it implausible that such underlying secular trends in lifestyle are a major contributor in the reported associations, we have now conducted additional analyses, excluding individuals with alcoholic cirrhosis of liver (ICD-10 code: K70.3) or COPD (chronic obstructive pulmonary disease, ICD-10 code: J44) as a proxy for heavy drinking or smoking. The results remained virtually unchanged.

   We have now added reasons for stratified analysis by calendar years in Method (Pages 8-9), and as below:

   “We performed subgroup analyses by sex, age at index date (<50, 50-60, or >60 years), age at follow-up (<60 or ≥60 years), history of somatic diseases (no or yes), and family history of psychiatric disorder (no or yes). We also performed subgroup analysis by calendar year at index date (1987-1996, 1997-2006, or 2007-2016) to check for potentially different associations over time (i.e., due to lifestyle factors that changed over time, including smoking and alcohol use).”

   We found somewhat higher risk of psychiatric disorder observed in recent calendar years than earlier years (as in shown Supplementary Table S3).

   We found similar associations between first-onset CVD and incident psychiatric disorder with and without excluding individuals with a history of alcoholic cirrhosis of liver or COPD, used as a proxy for heavy drinking or smoking. The table has now added as Supplementary Table S8, and also shown as below).

   We have now added justifications in Method (Page 10) and in Discussion (Page 21), and as below: In method, Page 10:

   “To account for potential impact of unmeasured confounding due to lifestyle factors, we performed a sensitivity analysis excluding individuals with a history of alcoholic cirrhosis of liver (ICD-10 code K703) or chronic obstructive pulmonary disease (COPD, ICD-10 code J44), as proxies for heavy drinking or smoking.”

   In Discussion (Page 21):
   “although we found similar results with and without excluding individuals with a history of liver cirrhosis or COPD, as proxies for heavy drinking or smoking (Supplementary Table S8). We did not have direct access to hazardous behaviors that could potentially modify this association, and therefore cannot exclude the possibility of residual confounding not fully controlled for in the sibling comparison.”

   References:

   1.Statista. https://www.statista.com/statistics/693505/per-capita-consumption-of-alcohol-in-thenordic-countries/. Retrieved on 19 Aug.
   2.Alcohol and Drug Report. Nordic Baltic Region. https://www.nordicalcohol.org/swedenconsumption-trends. Retrieved on 19 Aug. 3.Gunnarsdottir SA, Olsson R, Olafsson S, Cariglia N, Westin J, Thjódleifsson B, Björnsson E. Liver ;cirrhosis in Iceland and Sweden: incidence, aetiology and outcomes. Scandinavian journal of gastroenterology. 2009 Jan 1;44(8):984-93.

   Reviewer #2 (Public Review):

   Shen et. al investigated the associations between CVD and subsequent risk of psychiatric disorders using a prospective study design. The authors also performed subgroup analysis by sex, age at cohort entry and at follow-up, calendar year, history of somatic diseases, family history of psychiatric disease, and finally assessed the potential role of psychiatric comorbidity in cardiovascular mortality in CVD patients. The main takeaway of the analyses are the increased risk of psychiatric disorders in CVD patients compared to the different comparison groups.

   Though the study uses nationwide registers in a prospective study design setting, there are some methodological flaws with respect to study design.

   For assessing the primary aim the authors chose a rather unusual starting point by preselecting the exposure (CVD) group, rather than depicting the nationwide cohort of the general population followed up for a disease outcome with each category having exposed and unexposed individuals. Assuming that the population comparison group comes from the same study population as CVD patients, it is not clear why a similar strategy of study design as those cited in the manuscript (Zhang et. al 2015, Kivimäki et. al 2012, Godin et. al, 2012) was not followed. Similarly, one would expect sibling comparison group w.r.t outcome (psychiatric disorders) and not for exposure (CVD).

   Thank you for the comment. As correctly pointed out by the reviewer, we used a matched cohort design, both in the population- and sibling comparison. We firstly identified a nationwide cohort of general population who were born after 1932 and were residing in Sweden 1987-2016. We then identified all exposed individuals with first-ever diagnosis of CVD and matched population controls from this same nationwide population.

   A matched cohort design is applied here due to the strong confounding effects of some variables, e.g., age and sex, on the studied association between CVD and risk of psychiatric disorder. Exact matching on age and sex in our study makes the exposed and unexposed groups comparable and relief the confounding effects from matching factors in the design phase. Another practical viewpoint for why we use a matched cohort is a straightforward understanding of the comparison between exposed and unexposed groups being always at the same time, providing measures (such as risks and rates) during the follow-up period that are easily interpreted. Further, we have used this matched cohort design in many of our previous works [1,2] to maintain an identical design in both sibling and population comparison, so that the point estimates can be directly compared. The matched cohort design generates results of equal validity of the more conventional cohort design suggested by the reviewer [3] but has the additional quality of making the results from the various cohorts (here: population- and sibling comparison) more comparable. Our study therefore takes advantage of using a siblingcontrolled matched cohort, which is indeed a cohort design recommended for family-based studies [4] and provides results with similar validity as a full cohort.

   We have now added a sentence and a reference in Method to motivate the use of matched cohort design (Page 7).

   “We constructed a sibling-controlled matched cohort to control for the familial confounding according to guidelines for designing family-based studies.24”

   We have now updated the flowchart to add a box in the top reflecting the source population where both groups were identified from, shown in Supplementary Figure S1.

   References:

   1.Song H, Fang F, Arnberg FK, Mataix-Cols D, Fernández de la Cruz L, Almqvist C, Fall K, Lichtenstein P, Thorgeirsson G, Valdimarsdóttir UA. Stress related disorders and risk of cardiovascular disease: population based, sibling controlled cohort study. BMJ. 2019 Apr 10;365:l1255.
   2.Song H, Fang F, Tomasson G, Arnberg FK, Mataix-Cols D, Fernández de la Cruz L, Almqvist C, Fall K, Valdimarsdóttir UA. Association of Stress-Related Disorders With Subsequent Autoimmune Disease. JAMA. 2018 Jun 19;319(23):2388-2400.
   3.Sjölander A, Greenland S. Ignoring the matching variables in cohort studies–when is it valid and why?. Statistics in medicine. 2013 Nov 30;32(27):4696-708. 4.D'Onofrio BM, Lahey BB, Turkheimer E, Lichtenstein P. Critical need for family-based, quasiexperimental designs in integrating genetic and social science research. Am J Public Health. 2013 Oct;103 Suppl 1(Suppl 1):S46-55.

   Reviewer #3 (Public Review):

   Shen et al. investigated the relationship between the diagnosis of cardiovascular disease (CVD) and subsequent diagnosis of psychiatric disorders using national databases and health records over a 30year period in Sweden. They also investigated the association between the diagnosis of psychiatric disorder and subsequent CVD-related mortality. Comparisons were made between participants diagnosed with CVD and siblings without CVD, and between the CVD participants and random age- and sex-matched controls from the general population.

   They show that diagnosis of all types of CVD investigated was associated with increased risk of all types of psychiatric disorders considered, both in comparison to non-CVD siblings and general population controls. They also showed that diagnosis of psychiatric diagnosis subsequent to CVD diagnosis was associated with greater CVD-related mortality.

   A key strength of this study is the use of national databases and populations, as it has allowed for sufficiently large numbers for important subgroup analyses investigating specific types of CVD and psychiatric disorders. In addition to disease and disorder subtypes, the authors have investigated many other factors that may be important for understanding these relationships, including time of diagnosis during follow-up, year of diagnosis, age of participant, and various comorbidities. The duration of follow-up is another important strength of this study, as is the use of sibling controls to mitigate the potential confounding effect of genetic and early-life environment.

   However, while it is acknowledged as a limitation by authors, the lack of lifestyle data is a notable weakness of the study. The authors allude to causal inference in the abstract and discuss controlling for important confounding factors, but this is somewhat undermined by not being able to account for lifestyle factors, particularly since there are shared biological pathways such as inflammation linked to both CVD and many psychiatric disorders. As such, the associations reported in this study are potentially influenced substantially by unmeasured confounding related to lifestyle factors.

   Overall, this is important data, and the conclusions around these findings supporting surveillance of psychiatric disorders in individuals diagnosed with CVD due to its association with increased risk of mortality may be of interest to clinical settings.

   Thank you for the very positive comments.

   Read the original source
3. 

   eLife

   Sep 26, 2022

   Evaluation Summary:

   This is a useful quantification of the links of vascular disease on the development of subsequent mental health issues. It uses a robust dataset to quantify this association. Further work to focus the analyses, ensure claims are supported by the data, and consider alternative explanations is needed.

   (This preprint has been reviewed by eLife. We include the public reviews from the reviewers here; the authors also receive private feedback with suggested changes to the manuscript. Reviewer #2 agreed to share their name with the authors.)

   Read the original source
4. 

   eLife

   Sep 26, 2022

   Reviewer #1 (Public Review):

   This is a well-done analysis using the very robust Swedish national population registry.

   The study strengths include large size, prolonged follow-up, and use of two comparison populations.

   The main limitations which need to be addressed by the authors are accounting for reverse causality, namely if a psychiatric illness (PI) developed before or about the same time as the CVD. The much steeper risk relationships early after a CVD event are so suggestive. Some further analyses to tease out those with clearly NO PI before CVD would be in order.

   Second, for the robust matched cohort design, the authors age and sex matched each patient with 10 individuals from the general population and then also stratified their model by the matching variables. Could adjusting for matched factors in such cohort studies …

   Reviewer #1 (Public Review):

   This is a well-done analysis using the very robust Swedish national population registry.

   The study strengths include large size, prolonged follow-up, and use of two comparison populations.

   The main limitations which need to be addressed by the authors are accounting for reverse causality, namely if a psychiatric illness (PI) developed before or about the same time as the CVD. The much steeper risk relationships early after a CVD event are so suggestive. Some further analyses to tease out those with clearly NO PI before CVD would be in order.

   Second, for the robust matched cohort design, the authors age and sex matched each patient with 10 individuals from the general population and then also stratified their model by the matching variables. Could adjusting for matched factors in such cohort studies re-introduce bias into these estimates?

   Third, the range of PIs associated with CVD is a lot broader than would be expected or unexpected (eg eating disorders!).

   Lastly, the authors should try to account for secular changes in smoking and alcohol consumption or BMI over the study period. In particular, while Sweden never had very high smoking rates (due to Snus) alcohol use within specific cohorts might have both affected CVD risk (particularly stroke) and PI risk. Examining trends in for example liver cirrhosis over the study time period might help or use sales/consumption data. The authors do recognize a limitation in being unable to adjust for smoking, alcohol, and adiposity.

   Some additional analyses to address these points and some more caution in the discussion are required.

   Read the original source
5. 

   eLife

   Sep 26, 2022

   Reviewer #2 (Public Review):

   Shen et. al investigated the associations between CVD and subsequent risk of psychiatric disorders using a prospective study design. The authors also performed subgroup analysis by sex, age at cohort entry and at follow-up, calendar year, history of somatic diseases, family history of psychiatric disease, and finally assessed the potential role of psychiatric comorbidity in cardiovascular mortality in CVD patients. The main takeaway of the analyses are the increased risk of psychiatric disorders in CVD patients compared to the different comparison groups.

   Though the study uses nationwide registers in a prospective study design setting, there are some methodological flaws with respect to study design.
   For assessing the primary aim the authors chose a rather unusual starting point by preselecting the exposure …

   Reviewer #2 (Public Review):

   Shen et. al investigated the associations between CVD and subsequent risk of psychiatric disorders using a prospective study design. The authors also performed subgroup analysis by sex, age at cohort entry and at follow-up, calendar year, history of somatic diseases, family history of psychiatric disease, and finally assessed the potential role of psychiatric comorbidity in cardiovascular mortality in CVD patients. The main takeaway of the analyses are the increased risk of psychiatric disorders in CVD patients compared to the different comparison groups.

   Though the study uses nationwide registers in a prospective study design setting, there are some methodological flaws with respect to study design.
   For assessing the primary aim the authors chose a rather unusual starting point by preselecting the exposure (CVD) group, rather than depicting the nationwide cohort of the general population followed up for a disease outcome with each category having exposed and unexposed individuals. Assuming that the population comparison group comes from the same study population as CVD patients, it is not clear why a similar strategy of study design as those cited in the manuscript (Zhang et. al 2015, Kivimäki et. al 2012, Godin et. al, 2012) was not followed. Similarly, one would expect sibling comparison group w.r.t outcome (psychiatric disorders) and not for exposure (CVD).

   Read the original source
6. 

   eLife

   Sep 26, 2022

   Reviewer #3 (Public Review):

   Shen et al. investigated the relationship between the diagnosis of cardiovascular disease (CVD) and subsequent diagnosis of psychiatric disorders using national databases and health records over a 30-year period in Sweden. They also investigated the association between the diagnosis of psychiatric disorder and subsequent CVD-related mortality. Comparisons were made between participants diagnosed with CVD and siblings without CVD, and between the CVD participants and random age- and sex-matched controls from the general population.

   They show that diagnosis of all types of CVD investigated was associated with increased risk of all types of psychiatric disorders considered, both in comparison to non-CVD siblings and general population controls. They also showed that diagnosis of psychiatric diagnosis subsequent …

   Reviewer #3 (Public Review):

   Shen et al. investigated the relationship between the diagnosis of cardiovascular disease (CVD) and subsequent diagnosis of psychiatric disorders using national databases and health records over a 30-year period in Sweden. They also investigated the association between the diagnosis of psychiatric disorder and subsequent CVD-related mortality. Comparisons were made between participants diagnosed with CVD and siblings without CVD, and between the CVD participants and random age- and sex-matched controls from the general population.

   They show that diagnosis of all types of CVD investigated was associated with increased risk of all types of psychiatric disorders considered, both in comparison to non-CVD siblings and general population controls. They also showed that diagnosis of psychiatric diagnosis subsequent to CVD diagnosis was associated with greater CVD-related mortality.

   A key strength of this study is the use of national databases and populations, as it has allowed for sufficiently large numbers for important subgroup analyses investigating specific types of CVD and psychiatric disorders. In addition to disease and disorder subtypes, the authors have investigated many other factors that may be important for understanding these relationships, including time of diagnosis during follow-up, year of diagnosis, age of participant, and various comorbidities. The duration of follow-up is another important strength of this study, as is the use of sibling controls to mitigate the potential confounding effect of genetic and early-life environment.

   However, while it is acknowledged as a limitation by authors, the lack of lifestyle data is a notable weakness of the study. The authors allude to causal inference in the abstract and discuss controlling for important confounding factors, but this is somewhat undermined by not being able to account for lifestyle factors, particularly since there are shared biological pathways such as inflammation linked to both CVD and many psychiatric disorders. As such, the associations reported in this study are potentially influenced substantially by unmeasured confounding related to lifestyle factors.

   Overall, this is important data, and the conclusions around these findings supporting surveillance of psychiatric disorders in individuals diagnosed with CVD due to its association with increased risk of mortality may be of interest to clinical settings.

   Read the original source
7. 

   Version published to 10.1101/2022.06.27.22276962v1 on medRxiv

   Jun 27, 2022