Study Setting

The study consisted of three parts: an observational cohort study with the UK Biobank data, a summary-level MR analysis using the previously published genetic instrument and outcome GWAS summary statistics for kidney function traits using CKDGen, and an allele score–based MR within the UK Biobank (Figure 1). The aim of the first observational analysis was to yield a supportive individual-level observational association between psychologic wellbeing and CKD. The causal effects of dimensions of psychologic wellbeing and risk of CKD were investigated in the summary-level MR analysis. Allele score–based MR was performed to test the causal effects in an independent cohort.

Open in a separate window

Figure 1.

Study flow diagram. The study consisted of three parts: an observational analysis, a summary-level MR analysis with the CKDGen GWAS meta-analysis for CKD and eGFR traits, and an allele score–based MR with individual-level data in the UK Biobank.

UK Biobank and the Observational Study Data

The UK Biobank is a population-scale prospective cohort including participants aged 40–69 years. The study recruited >500,000 participants from 2006 to 2010 from 22 assessment centers in the UK. The individuals included in the UK Biobank cohort were healthier than the general population in the UK, had a low prevalence of CKD, and had some differences in characteristics, when compared with the general population. The other details of the UK Biobank project have been published previously.^14,15

For the observational investigation, among the 502,505 UK Biobank participants available in this study, we included 468,521 individuals with identifiable baseline eGFR values, calculated by the CKD Epidemiology Collaboration equation on the basis of serum creatinine–cystatin C values, measured by the standardized enzymatic method.

Phenotypic Exposures of the Observational Investigation

The exposures of the observational analysis were selected to reflect similar dimensions of psychologic wellbeing as the available genetic instruments: general happiness (to reflect positive affect in the genetic instrument), belief their own life is meaningful (to reflect life satisfaction in the genetic instrument), broad depression (to reflect depressive symptoms in the genetic instrument),^16,17 and neuroticism (to reflect neuroticism in the genetic instrument). Responses to questions about general happiness (graded using six possible responses from “extremely happy” to “extremely unhappy”) and belief that one’s own life is meaningful (graded using five possible responses from “an extreme amount” to “not at all”) were collected using a touchscreen questionnaire at the baseline visit. The presence of broad depression was determined by a previous history of medical visits for psychiatric symptoms and has been utilized in the literature on depression in the UK Biobank, because it has been reported to be the most genetically trackable phenotype in the database. The neuroticism score was on the basis of 12 questions related to neurotic symptoms (e.g., mood swings, nervous feelings, tenseness), and the number of “yes” answers consisted of the sum score. The number of individuals reporting the exposures of the observational investigation were different according to the implemented dimensions of psychologic wellbeing in the observational investigation: general happiness (n=146,887), belief their own life is meaningful (n=144,186), broad depression (n=468,521), and neuroticism score (n=375,500). Those without exposure information were not included in the according analysis with the exposure. The details of the collected data, including covariates and missing information, are described in the Supplemental Methods.

CKD Outcome in the UK Biobank Data

Considering the relatively low prevalence of CKD compared with the general population in the UK Biobank,^14,18 we implemented a broad definition to define CKD stage 3–5 in the UK Biobank data. The broad definition was to avoid possible underestimation of CKD events because baseline eGFR values could not capture CKD events that occur after the baseline visit, although both prevalent and incident events can be assessed as outcomes in an MR analysis.¹⁹ The CKD event was determined by presence of one or more of the following histories: baseline eGFR <60 ml/min per 1.73 m² by creatinine-, cystatin C–, or creatinine-cystatin C–based CKD Epidemiology Collaboration equation; a prevalent/incident event of ESKD; or an International Classification of Diseases 10^th revision diagnostic code indicative of CKD stage 3–5, dialysis, or transplantation.

Statistical Methods for Observational Investigation

The observational association between exposures and CKD stage 3–5 was investigated by the logistic regression model. The first model was age and sex adjusted, and the second multivariable model was additionally adjusted for important confounders, including hypertension, diabetes, body mass index, and the number of household members. A more stringently adjusted model was not constructed, as observed psychologic wellbeing senses would be a sum of various medical and social factors. All analyses were performed by complete-case methods. All statistical analyses were performed using R (version 3.6.2, the R Foundation), and two-sided P values <0.05 were considered significant for the observational investigations and in the genetic analysis, unless otherwise specified.

Genetic Instrument for MR Analysis

The genetic instrument was developed from a novel model-averaging genome-wide association meta-analysis revealing independent (r ²<0.1 in 250 kb window) single nucleotide polymorphisms (SNPs) with genome-wide significance (P<5×10⁻⁸) for standardized phenotypes of positive affect (191 SNPs), life satisfaction (148 SNPs), depressive symptoms (239 SNPs), and neuroticism (263 SNPs) from a previous study.²⁰ The study analyzed genetic information from the Social Science Genetic Association Consortium (SSGAC),²¹ Understanding Society,²² UK Biobank, 23andMe,^23,24 and CHARGE²⁵ cohorts of European ancestry, the details of which are presented in Supplemental Table 1. The model-averaging genome-wide association meta-analysis was performed with the published summary statistics, and the joint model averaged effect sizes were calculated, combining the multiple phenotypes as z scores setting the variance of each phenotype in the summary statistics to 1. Polygenic prediction showed that the identified SNPs explained certain variance for positive affect (r ²=1.1%), life satisfaction (r ²=0.9%), depressive symptoms (r ²=1.6%), and neuroticism (r ²=1.6%) in the subsamples. The study reported that related genes were differentially expressed in the subiculum and GABAergic interneurons. We redirected the effects of the genetic instrument toward positive wellbeing sense, and the effect sizes for depressive symptoms/neuroticism were negatively correlated with those for positive affect/life satisfaction.

To construct the genetic instruments for psychologic wellbeing sense and perform MR investigation, we considered the three assumptions of MR analysis required to demonstrate valid causal effects.¹⁰ First, the relevance assumption means the genetic instrument is strongly associated with the exposure of interest. The previous GWAS meta-analysis results showed SNPs were strongly associated with wellbeing exposure, and we applied a more conservative approach, including SNPs with a Bonferroni-adjusted significance level (P<0.05/5,000,000 [1×10⁻⁸]) of association, with each wellbeing dimension serving as the main genetic instrument. To assess the second assumption, the independence assumption, we trimmed the genetic instruments by excluding SNPs that showed a strong association with possible major confounders, including hypertension, diabetes, obesity, income grade, and number of household members. For this, the UK Biobank dataset for the genetic analysis, described below, was utilized because the data included information on the possible confounders. A GWAS was performed for the possible confounders with SNPs included in the genetic instrument, with adjustment for age, sex, age×sex, age², and the first 20 principal components, and the SNPs with an association with any of the possible confounders that reached a P value <1×10⁻⁵ were excluded from the genetic instrument.¹¹ Analytically, additional sensitivity MR analyses were performed to investigate pleiotropy-robust causal estimates. The third assumption, the exclusion restriction assumption, means the genetic effect should be through the exposure of interest. Although this assumption cannot be formally tested, we performed median-based sensitivity MR analysis, which can relax this assumption in certain portions of the instruments.²⁶

Furthermore, we again pruned the genetic variants within a 500 kb window to include independent signals (r ²<0.1) by the genome data of the UK Biobank. Steiger filtering was performed to ensure the direction of the genetic effects was due to the exposure of interest toward the outcome.²⁷ Last, the SNPs present in both summary statistics for the genetic instrument and for the CKDGen results were included, and the SNPs that were palindromic with intermediate allele frequencies were disregarded.

CKDGen GWAS Meta-analysis for Summary-level MR Analysis

The CKDGen GWAS meta-analysis is the largest GWAS for kidney function outcomes to date.²⁸ We implemented the summary statistics for log-transformed eGFR (calculated from creatinine values, log ml/min per 1.73 m²) and CKD (eGFR <60 ml/min per 1.73 m²) from individuals of European ancestry (n=567,460). The individuals of European ancestry in the CKDGen had a median age of 54 years old, and 50% of them were male, with a median eGFR of 91.4 ml/min per 1.73 m² and a prevalence of CKD of 9% (Supplemental Table 2). There was possible partial overlap in the samples in the CKDGen data with the GWAS that provided the genetic instrument (e.g., Atherosclerosis Risk in Communities Study).

Summary-level MR Methods

First, as we stated above, we utilized the SNPs associated with psychologic wellbeing exposures that reached a Bonferroni-corrected significance level (P<1×10⁻⁸). Next, we filtered SNPs with stronger associations (P<1×10⁻¹⁰) with the exposures to decrease the possibility of remaining horizontal pleiotropy, and repeated the analysis. The main summary-level MR method was the multiplicative random-effects inverse variance weighted method.²⁹ As the level of heterogeneity of the utilized variants was significant, the method yielded the same results as the random-effects inverse variance weighted method, which allows balanced pleiotropy. For the sensitivity MR analysis, MR-Egger regression, which yields pleiotropy-robust causal estimates, was performed.³⁰ The analysis was accompanied by the MR-Egger test for directional pleiotropy, also known as the MR-Egger intercept, and a P value <0.1 was considered to indicate the possibility of directional pleiotropy, namely, a nonzero MR-Egger intercept. Nonsignificant evidence for directional pleiotropy or a significant result (P<0.05) from MR-Egger regression in patients with suspected pleiotropy was necessary to indicate a significant causal estimate, along with a significant result from the inverse variance weighted method. Another sensitivity analysis was performed by Mendelian Randomization Pleiotropy ReSidual Sum and Outlier (MR-PRESSO).³¹ MR-PRESSO can detect and correct the effects caused by outliers and was performed when the MR-PRESSO global test indicated the presence of significant outliers. In addition, median-based methods were implemented, which have strength when invalid instruments are present. The simple median method, providing valid causal estimates while allowing up to 50% invalid instruments, and the weighted median method, allowing up to 50% of invalid weights contributed by the genetic instruments, were utilized.²⁶ Cochran’s Q statistics P values were calculated to assess the presence of heterogeneity among the implemented genetic instruments. Regarding eGFR outcomes, the log-transformed eGFR unit was changed to a % change unit. The summary-level MR analysis was performed by the TwoSampleMR package in R (version 3.6.2, the R foundation,³²), and statistical codes to call the causal estimates by the package are provided in the Supplemental Methods.

Sensitivity Analysis for Summary-level MR

It has been reported that the presence of overlapping samples between the data for genetic instrument generation and outcome assessment may cause bias toward observational associations. Thus, we performed an additional sensitivity analysis, which had been previously suggested,³³ by reconstructing the genetic instrument including SNPs with different association strengths with the exposure trait of interest. This method has been suggested to inspect whether the effects from overlapping samples may alter the main findings by testing the effects of changing the instrumental power of genetic instruments. We filtered SNPs within a range of thresholds (P<5×10⁻⁸ to <1×10⁻¹⁰), repetitively performed summary-level MR analysis by the inverse variance weighted method, and performed the MR-Egger test to detect directional pleiotropy.

UK Biobank Participants for Allele Score–based MR

For allele score–based MR,³⁴ 337,138 participants of White British ancestry passed the quality control filter. Those who were outliers in terms of heterozygosity or had a missing rate, those with sex chromosome aneuploidy and those with excess kinship (≥10 third-degree relatives) were excluded on the basis of the information determined by the UK Biobank consortium, and the samples included in the genetic principal component calculation were from unrelated individuals.³⁵ In addition, those with missing creatinine or cystatin C values were excluded, because the information was necessary to determine the study outcome, and the remaining 321,024 participants were finally included in the investigation.

Statistical Methods for Allele Score–based MR

The UK Biobank data were the largest single cohort used to validate the findings of summary-level MR with individual-level data. Yet, a possible bias from clinical confounders was present, as the UK Biobank sample overlapped with the samples included in the study for genetic instrument development.³³ However, assessment of causal estimates in different individual-level data has been performed in previous MR studies for validation purposes, and the analysis had strength as additional consideration of individual-level covariates was possible.^19,36,37

We calculated allele scores for the exposures by multiplying the gene dosage matrix with the effect sizes of the main genetic instrument for positive affect, life satisfaction, depressive symptoms, and neuroticism, respectively, using PLINK 2.0 (version alpha 2.3).³⁸ As allele scores were calculated from summing the effect size betas, one increase in the allele score reflected one z score increase in the previous model-averaging genome-wide association meta-analysis results setting phenotype variance to 1 as the summary-level MR.³⁷ The association between genetically predicted psychologic wellbeing sense and log-transformed creatinine–cystatin C eGFR or CKD was tested by linear or logistic regression with adjustment for age, sex, genotype measurement batch, and the first 10 principal components. We additionally performed an analysis by adding phenotypical hypertension, diabetes mellitus, body mass index, number of household members, and income grades in the regression model.³⁷ The effect sizes for the eGFR outcome were transformed into % change units as the summary-level MR analysis.

Observational Association Between Psychologic Wellbeing and CKD

Patients’ age and sex were strongly associated with their psychologic wellbeing (Supplemental Table 4). In the age- and sex-adjusted model (Table 1), those who reported general happiness status showed approximately 35% lower odds for CKD stage 3–5 when compared with those who were unhappy. The results remained similar even after adjusting for body mass index, hypertension, diabetes, and the number of household members. Those who considered their own life to be “very” or “extremely” meaningful showed approximately 17% lower odds for CKD in the multivariable model than the others. Those who reported broad depression or neuroticism scores ≥5 had approximately 11% or 13% higher adjusted odds for CKD in the multivariable model than those with better psychologic wellbeing.

Table 1.

Observational association between psychologic wellbeing and odds of CKD stage 3–5 in the UK Biobank

Exposure	N with Exposure	N with CKD and Proportion	Age,- Sex-Adjusted Model		Multivariable Model
			Adjusted OR (95% CI)	P	adjusted OR (95% CI)	P
General happiness
Moderately, very, or extremely happy	138,686	5404 (3.9%)	0.65 (0.58 to 0.73)	<0.001	0.71 (0.63 to 0.80)	<0.001
Moderately, very, or extremely unhappy	8201	347 (4.2%)	Reference		reference
Belief that own life is meaningful
Very or an extreme amount	94,812	3601 (3.8%)	0.83 (0.78 to 0.88)	<0.001	0.87 (0.82 to 0.92)	<0.001
A moderate amount, a little, or not at all	49,374	2034 (4.1%)	reference		reference
Broad depression
Presence of broad depression	161,349	10,484 (6.5%)	1.22 (1.19 to 1.26)	<0.001	1.11 (1.09 to 1.15)	<0.001
Absence of broad depression	307,172	18,090 (5.9%)	reference		reference
Neuroticism
Neuroticism score ≥5	153,918	9137 (5.9%)	1.21 (1.17 to 1.24)	<0.001	1.13 (1.10 to 1.16)	<0.001
Neuroticism score <5	221,582	12,882 (5.8%)	reference		reference

Open in a separate window

The multivariable model was adjusted for age, sex, body mass index, hypertension, diabetes mellitus, and number of household members. The complete-case multivariable analysis was performed for 145,443 individuals with general happiness exposure (5663 patients with CKD), 142,784 individuals with meaningful life exposure (5549 patients with CKD), 458,965 individuals with broad depression exposure (27,631 patients with CKD), and 369,575 individuals with neuroticism exposure (21,363 patients with CKD), respectively, without any missing information in the covariates. OR, odds ratio; CI, confidence interval.

Genetic Instrument for Psychologic Wellbeing Spectra

After excluding possible confounder-associated SNPs (Supplemental Table 5), the independent SNPs (500-kb window, r ² < 0.1) with Bonferroni-corrected significant level association (P<1×10⁻⁸) with each psychologic wellbeing dimension were filtered in. The summary statistics of the main genetic instrument for positive affect (102 SNPs), life satisfaction (76 SNPs), depressive symptoms (125 SNPs), and neuroticism (133 SNPs) are presented in Supplemental Table 6.

Summary-level MR Results

In the summary-level MR (Figure 2, Table 2, Supplemental Figure 1) conducted with the inverse variance weighted method, genetically predicted positive affect was associated with a lower risk of CKD. In contrast, depressive symptoms showed significant causal estimates toward a higher risk for CKD. With the SNPs and Bonferroni-corrected association level significance (P<1×10⁻⁸), the causal estimates from depressive symptoms to CKD risk were supported by the utilized sensitivity analysis results, including the MR-Egger, weighted median, simple median, and MR-PRESSO results. However, only the MR-PRESSO analysis yielded significant causal estimates from positive affect to a lower risk for CKD. Otherwise, the causal estimates from life satisfaction or neuroticism were found to be nonsignificant by the inverse variance weighted method.

Open in a separate window

Figure 2.

MR analysis results indicated the causal linkage between dimensions of psychologic wellbeing and kidney function. The summary statistics for eGFR and CKD outcomes were introduced from the CKDGen GWAS meta-analysis. The effect sizes in the summary-level MR were from the genetic predisposition to a z score in the previous model-averaging genome-wide association meta-analysis setting phenotype variance to 1. The presented causal estimates from the summary-level MR analysis are from multiplicative random-effect inverse variance weighted method. The effect sizes of the allele score–based MR was similarly scaled to sum of the betas with the unit of a z score in the previous model-averaging genome-wide association meta-analysis setting phenotype variance to 1. The allele score–based MR analysis results are from the covariate-adjusted logistic regression model including age, sex, the first 10 genetic principal components, body mass index, hypertension, diabetes mellitus, income grade, and number of household members. The unit of the eGFR outcome was % change, calculated from the log-transformed eGFR values. Cr, creatinine; CI, confidence interval; OR, odds ratio.

When we filtered the SNPs with stronger associations (P<1×10⁻¹⁰) with the exposures (Supplemental Table 7), the inverse variance weighted method again yielded significant causal estimates from positive affect to a lower risk for CKD, and from depressive symptoms to a higher risk for CKD. The simple-median, weighted-median, and MR-PRESSO results were significant, as the main findings for genetically predicted positive affect being associated with a lower risk for CKD. Although the causal estimates by MR-Egger regression did not reach a statistically significant level, the MR-Egger intercept P value indicated the absence of significant directional pleiotropy, supporting the causal estimates from the inverse variance weighted method. In this analysis, causal estimates from depressive symptoms to a higher risk for CKD were supported by median-based methods, with the results of the MR-Egger test for directional pleiotropy indicating the absence of a significant pleiotropic effect.

For eGFR outcome (Figure 2, Table 3, and Supplemental Figure 1), the inverse variance weighted method showed genetically predicted positive affect to be associated with higher eGFR values. However, the pleiotropy-robust sensitivity MR analysis results did not support the finding as the MR-PRESSO results did. Genetically predicted depressive symptoms were significantly associated with lower eGFR values by the inverse variance weighted, MR-Egger regression, and MR-PRESSO methods. Again, the causal estimates from life satisfaction or neuroticism did not reach a significant level.

When the SNPs with higher association strength (P<1×10⁻¹⁰) were utilized (Supplemental Table 8), the causal estimates from depressive symptoms to a lower eGFR values were significant according to the inverse variance weighted method, as supported by the median-based and MR-PRESSO results. Although the MR-Egger regression result did not reach a significant level, the absence of directional pleiotropy was supported by the MR-Egger intercept P value. For the SNPs with P<1×10⁻¹⁰ for the association for each type of wellbeing, genetically predicted positive affect was no longer significantly associated with a higher eGFR, and the findings for life satisfaction or neuroticism exposure remained to be nonsignificant.

Additional Sensitivity Analysis for Summary-level MR

In an additional sensitivity analysis, the causal estimates made by the summary-level MR showed similar findings as the main results, even after ranges of P value thresholds were applied for wellbeing sense to define the genetic instruments (Supplemental Table 9). The findings suggested that genetically predicted depressive symptoms were associated with a higher risk of CKD or lower eGFR, and genetically predicted positive symptoms were associated with a lower risk of CKD when the conventional genome-wide significance level (P<5×10⁻⁸) or a different threshold P<1×10⁻⁹ was applied and the number of variants utilized as the genetic instruments was changed.

Allele Score–based MR Results

In the allele score–based MR in the UK Biobank, genetically predicted positive affect or life satisfaction was significantly associated with higher eGFR change (Table 4 and Figure 2). In contrast, genetically predicted depressive symptoms or neuroticism were significantly associated with eGFR decrease. The results remained significant even when we additionally adjusted clinical covariates including hypertension, diabetes, body mass index, income grades, and number of household members.

For CKD outcome, genetically predicted positive affect was significantly associated with lower risk of CKD, whereas other three wellbeing senses did not show significant association. After adjusting clinical covariates, genetically predicted positive affect and life satisfaction was significantly associated with lower risk of CKD, whereas higher allele scores for neuroticism were associated with higher risks of CKD. In contrast, genetically predicted depressive symptoms did not show a significant association with CKD risks, although the direction of the association from a higher allele score for depressive symptoms was toward a higher risk of CKD.

The study was on the basis of the data provided by the UK Biobank consortium (application 53799). We thank the investigators of the CKDGen, BIOS, and SSGAC consortia who provided the summary statistics for the genetic instrument and the outcome of this study.

The corresponding author attests that all listed authors meet the authorship criteria and that no others meeting the criteria have been omitted. D.K. Kim, H. Lee, K. Kim, K.W. Joo, and S. Park contributed to the conception and design of the study; C.S. Lim, D.K. Kim, J.P. Lee, K.W. Joo, M.W. Kang, S.S. Han, S. Lee, Y. Kim, Y.C. Kim, Y. Lee, and Y.S. Kim provided statistical advice and interpreted the data; K. Kim and S. Park performed the main statistical analysis, assisted by S. Lee and Y. Kim; C.S. Lim, D.K. Kim, H. Lee, J.P. Lee, K.W. Joo, and Y.S. Kim, and provided advice regarding the data interpretation; C.S. Lim, H. Lee, J.P. Lee, K.W. Joo, S.S. Han, Y.C. Kim, and Y.S. Kim provided material support during the study; D.K. Kim and S. Park had full access to all data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis; All authors participated in drafting the manuscript, reviewed the manuscript, and approved the final version to be published.