Study design and population

Data from two different population-based prospective cohort studies in the Netherlands were used: VAccination Study COrona (VASCO) [10] and PIENTER Corona (PICO) [11, 12]. Both cohorts were sampled from the general population and asked to complete the Dutch translation of the 12-item Short-Form Health Survey (SF-12) version 1. We analysed data from both cohorts to improve robustness and validity of our results and to cover a longer observation period. Participants from both cohorts aged between 18 and 85 years, and who completed the SF-12 questionnaire at least once, were included. Both studies were conducted in accordance with the principles of the Declaration of Helsinki. The VASCO study protocol was approved by the not-for-profit independent Medical Ethics Committee of the Stichting Beoordeling Ethiek Biomedisch Onderzoek (BEBO), Assen, the Netherlands (Clinical Trial Registration NL9279). The PICO study protocol was approved by the Medical Ethics Committee MEC-U, the Netherlands (Clinical Trial Registration NTR8473). All participants provided written informed consent.

VASCO is an ongoing population-based prospective cohort study, in which 45,553 community-dwelling persons aged 18–85 years were enrolled between 3 May 2021 and 15 December 2021 [10]. Participants were recruited through random sampling of the Dutch population register and campaigns on social media and traditional media. Participants had to be able to understand Dutch, as all study materials were written in Dutch, and were included irrespective of their COVID-19 vaccination status or intention to get vaccinated. The main aim of this 5-year follow-up study is to estimate vaccine effectiveness of COVID-19 vaccination against SARS-CoV-2 infection over time. At baseline, data on sociodemographic variables and health status were collected. Next to questions regarding SARS-CoV-2 testing and COVID-19 vaccination used for estimating vaccine effectiveness, participants were asked to complete the SF-12 questionnaire to assess their HRQoL at baseline and every three months. Since participants entered the VASCO cohort over a period of 6 months and the SF-12 questionnaire is administered every three months, continuous data are available from 3 May 2021 until 31 July 2022.

PICO is a nationwide, population-based seroepidemiological cohort study investigating SARS-CoV-2 antibodies and risk factors among the Dutch population as its primary aim. Participants were initially sampled in March 2020 from the in 2016/2017 established PIENTER-3 serosurvey cohort in which participants were invited via a two-stage cluster design [11, 12]. To enhance countrywide geographical coverage and maintain power, the PICO study population was supplemented twice (in June 2020 and November 2021) by actively inviting additional persons randomly selected from the Dutch population registry [13]. This resulted in an overall study population of 11,159 participants aged 1–92 years. In each round, participants completed a questionnaire including questions on sociodemographic factors, health status, COVID-19 vaccination, and SARS-CoV-2-related symptoms. Study rounds from February 2021 onwards additionally included the SF-12 questionnaire to assess HRQoL for participants aged 15 years and older. To keep study periods more concise, data of the first four weeks of each round (February, June, and November 2021) were included, covering the majority of the completed questionnaires (98.4%, 94.5% and 95.4%, respectively).

Outcome measures

The SF-12, an abbreviated version of the SF-36 HRQoL questionnaire, includes 12 items covering eight health concepts: physical functioning (PF), role limitations due to physical health problems (RP), bodily pain (BP), general health (GH), vitality (energy/fatigue) (VT), social functioning (SF), role limitations due to emotional problems (RE) and mental health (psychological distress and psychological wellbeing) (MH) [14]. These concepts are converted into a physical health (PCS) and mental health (MCS) component score. The PCS score primarily encompasses answers to PF, RP, BP and GH questions and the MCS score answers to VT, SF, RE and MH questions. PCS and MCS scores were weighted using standard regression coefficients obtained in the general US population [15, 16]. PCS and MCS scores range from 0 to 100 with higher scores denoting better health and a score of 50 indicating average health. Minimal clinically important differences (MCID) for PCS and MCS scores are suggested to range between 3 and 5 [17, 18], but may be somewhat outside this range in certain patient groups [19–22]. In addition to PCS and MCS scores, Short-Form Six Dimensions (SF-6D) health utility scores were derived from seven items from the SF-12 covering six of its health concepts (PF, role limitations (RE/RP), SF, BP, MH and VT) [23]. Scores range from 0.3 to 1 (UK population average 0.8), with 1 indicating full health [23, 24]. The MCID of the SF-6D is estimated to be 0.033 (95% CI: 0.029 to 0.037) with a range of 0.010 to 0.048 [25].

The SF-12 version 1 measured HRQoL related to mental health, physical health and health utility in the four weeks prior to completion. In VASCO and the online PICO questionnaire in the February 2021 round, participants received the SF-12 version 1 UK. The paper-based February 2021 questionnaire and all questionnaires of subsequent PICO rounds contained the SF-12 version 1 US. Results of the SF-12 version 1 UK were transformed to the SF-12 version 1 US following standard procedure and all questionnaires were analysed in the SF-12 version 1 US format [26, 27].

Covariates

As a proxy for the stringency of COVID-19 measures in the Netherlands, the Dutch Stringency Index of the Oxford Coronavirus Government Response Tracker (OxCGRT) was used [28]. The Stringency Index is a composite measure based on nine response indicators such as school closures, workplace closures, and travel bans, rescaled to a value from 0 to 100 (strictest), which records the strictness of government policies [28].

Four risk groups were defined combining age group (18–59 and 60–85 years) and the presence of a medical risk condition (yes/no). A medical risk condition was defined as reporting at least one of the following physician-diagnosed and/or drug-treated conditions: asthma, diabetes (type 1, type 2 and type unknown), asplenia, cardiovascular disease, immune deficiency, cancer (currently, treated and currently, untreated), liver disease, lung disease, neurological disease, kidney disease and organ or bone marrow transplantation.

Study period

Data between 10 February 2021 and 31 July 2022 were available from the two cohorts, and were analysed separately. A complete case analysis was performed. We defined three four-week time periods based on PICO study rounds in 2021 and an additional four-week period with few restrictive measures in 2022: (1) 10-02-2021–09-03-2021; (2) 14-06-2021–11-07-2021; (3) 01-11-2021–28-11-2021; (4) 30-05-2022–27-06-2022. The four time periods captured different stages of the epidemic curve as well as different phases of restrictive measures that were in place (Fig 1). In period one, incidence of infections was relatively low, but there was a strict lockdown including curfew and closure of non-essential shops. Incidence was low in period two, but the Delta variant of concern caused a peak in infections in period three. In both periods, restrictive measures were less intensive than in period one, with intensiveness decreasing in period two and increasing in period three. Period four captures a period in which there is a peak in infections caused by the Omicron BA.5 variant of concern (i.e., the third Omicron wave in the Netherlands), but most restrictive measures were lifted.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Timeline of OxCGRT’s Stringency Index and SARS-CoV-2 infection incidence* in the Netherlands during our study period with four predefined and separately studied periods colour-indicated.

* Dutch incidence data (Infectieradar) is available from: https://www.infectieradar.nl/results.

https://doi.org/10.1371/journal.pone.0300324.g001

Statistical analysis

Characteristics of the participants in both cohorts were presented using descriptive statistics to assess distribution of the prognostic factors age, sex, ethnic background, comorbidities, risk groups and educational level.

Primary objective: Mean PCS, MCS and SF-6D utility scores with standard deviations (SDs) during each of the four different periods were calculated cross-sectionally as well as averages over the four periods. Results were stratified by age, sex and medical risk condition. Per age group, scores between males and females and those with and without a medical risk condition were tested for significance using a two-sample t-test. P-values <0.05 were considered statistically significant. Stratification by sex excluded participants who classified their sex as other, as limited numbers were available. Additionally, distributions of scores of each SF-12 item were studied separately by age group.

Secondary objective: Mean weekly PCS, MCS and SF-6D utility scores with 95% confidence interval (CI) were calculated for VASCO participants over the period of 17 May 2021 to 31 July 2022. Questionnaires that were completed between 3 and 17 May 2021 (n = 5) were excluded because of limited inclusion numbers in these first two weeks. Results were stratified by risk group. PCS, MCS and SF-6D utility scores were modelled as a function of risk group and time using a generalized additive model from the R package mgcv (with penalized B-splines, 20 knots). For each outcome, Spearman correlations with the OxCGRT’s Stringency Index were calculated, with a lag time of two weeks representing the midpoint of the four weeks measured in the SF-12.

Multiple sensitivity analyses were performed. Firstly, all analyses were repeated using Dutch population weights [29] for PCS and MCS and crude sum scores of SF-12 items associated with mental or physical health [30, 31]. Secondly, for our primary objective we weighted the results of both cohorts to the general Dutch population based on age, sex, ethnicity and area of residence to ensure comparability between the two cohorts. Lastly, for our secondary objective, Spearman correlations were calculated using varying lag times of zero, one, three and four weeks. Additionally, we explored correlations for the pre-Omicron period (until 27 December 2021) and Omicron period (from 27 December 2021) with two-week lag-time separately, as the Omicron period marked the start of a new phase in the pandemic with high infection numbers but decreased individual disease burden and less restrictive measures.

Data cleaning was performed in SAS 94 M7 English for the PICO study and in R version 4.2.2 for the VASCO study. Statistical analyses of both studies were performed using R version 4.2.2.

Study population

A total study population of 53,432 participants (VASCO: 45,413; PICO: 8,019) was included (Table 1). 90% of VASCO participants completed at least three SF-12 questionnaires (range: 1–5). For PICO participants, 50% participated in all three PICO rounds since new study participants were recruited in the third period (November 2021). VASCO participants were older than PICO participants (median 60 vs 53 years) and the age distribution differed between the two studies with the 50–70 year age groups being overrepresented in VASCO. Additionally, VASCO participants were more often female (62.9% vs 57.3%) and highly educated (56.9% vs 48.4%). The distributions of ethnicities (89.5% vs 89.2% Dutch) and having at least one comorbidity (33.0% vs 32.6%) were comparable between the two studies. The largest VASCO subgroup was 18–59 year-olds without a medical risk condition (N = 17,326), followed by 60–85 year-olds without a medical risk condition (N = 14,701), 60–85 year-olds with a medical risk condition (N = 9,133), and 18–59 year-olds with a medical risk condition (N = 4,253). For PICO, the respective group sizes were 4,097, 1,701, 1,472 and 749. The percentages of participants with missing covariate data on age, sex and comorbidities was limited (VASCO: 0.3%; PICO: 0.2%).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Characteristics of all PICO and VASCO participants included in this study^a.

https://doi.org/10.1371/journal.pone.0300324.t001

Each of the four selected time periods included different numbers of participants (S1 File, Fig S1.1). In period one only PICO participants were included (N = 5,507). Period two and three consisted of both VASCO and PICO participants with 15,902 (VASCO: 11,102; PICO: 4,800) and 15,166 (VASCO: 8,407; PICO: 6,759) participants respectively. Period four included only VASCO participants (N = 12,837). Participants in period one were the youngest (median age 54 years), were least often female (56.5%) and least often had a medical indication (25.9%) compared to period two, three and four (median age 62, 58 and 61 years; females 60.5%, 61.5% and 62.0%; medical indication 30.3%, 32.0% and 29.6%, respectively).

HRQoL stratified by age group, sex and presence of medical risk condition

VASCO and PICO data showed comparable trends in SF-6D utility scores (Fig 2), MCS (Fig 3), and PCS (Fig 4) by age group and sex, by age group and presence of a medical risk condition, and over periods. Therefore, we describe the average results over all periods of PICO participants only (all results are presented in S1 File, Tables S1.1-S1.3). Using Dutch general population weights or crude sum scores of SF-12 items associated with mental or physical health did not affect observed patterns (S1 File, Figs S1.2-S1.7). Weighting of the results to the general Dutch population only marginally influenced the results. These results were therefore disregarded in the final analysis. Separate mental and physical health items showed similar patterns by age group and are therefore not presented individually.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2.

Health utility (SF-6D) in PICO and VASCO in four periods by age group and sex (A) and by age group and medical risk condition (B). Mean scores per study over the four periods are visualized by the red line.* Even though SF-6D scores can range from 0 to 1 (population average 0.8), y-axis ranges from 0.7 to 0.9 to better visualize differences.

https://doi.org/10.1371/journal.pone.0300324.g002

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3.

Mental health (MCS) in PICO and VASCO in four periods by age group and sex (A) and by age group and medical risk condition (B). Mean scores per study over the four periods are visualized by the red line.* Even though MCS scores can range from 0 to 100 (population average 50), y-axis ranges from 40 to 60 to better visualize differences.

https://doi.org/10.1371/journal.pone.0300324.g003

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4.

Physical health (PCS) in PICO and VASCO in four periods by age group and sex (A) and by age group and medical risk condition (B). Mean scores per study over the four periods are visualized by the red line.* Even though PCS scores can range from 0 to 100 (population average 50), y-axis ranges from 40 to 60 to better visualize differences.

https://doi.org/10.1371/journal.pone.0300324.g004

Mean SF-6D utility scores were lowest among 18–24 year-olds, increased with age up to 65–79 years, and declined thereafter (Fig 2). Between these age groups in PICO, mean male scores increased from 0.79 to 0.86 and then decreased to 0.83; mean scores of those without a medical risk condition increased from 0.77 to 0.87 and then decreased to 0.85 (S1 File, Table S1.1A). With significant differences observed for almost all age group (S1 File, Table S1.1), female participants scored lower than males (Fig 2A) and participants with a medical risk condition had lower scores than those without (Fig 2B), with differences up to 0.05.

Mean MCS scores were also lowest among the youngest age groups up to 30 years (Fig 3). Scores increased with age up to 65–79 years and plateaued thereafter. Between these age groups in PICO, mean male scores increased from 47 to 56 and mean scores of participants without a medical risk condition increased from 46 to 55 (S1 File, Table S1.2A). On average, for the MCS again, females had lower scores than males with differences up to 2 (Fig 3A), and participants without a medical risk condition scored higher compared to those with a medical risk condition with differences up to 4 (Fig 3B). For the majority of age groups these differences between males and females and those with and without a medical risk condition were significant (S1 File, Table S1.2).

Contrary to SF-6D utility and MCS scores, mean PCS scores were highest in the youngest age groups up to 30 years (Fig 4). PCS decreased gradually with age up to 70–79 years and decreased sharply thereafter. Between these ages in PICO, mean male scores decreased from 56 to 48 and mean scores of participants without a medical risk condition from 55 to 50 (S1 File, Table S1.3A). Also, for PCS, mostly significantly lower scores were observed for females compared to males with differences up to 3 (Fig 4A).Larger, strongly significant differences (up to 6; S1 File, Table S1.3) were observed by presence of a medical risk condition, with lower scores for those with a medical risk condition (Fig 4B).

HRQoL over time and correlation with stringency of restrictive measures

Mean weekly SF-6D utility, MCS and PCS scores in the VASCO cohort showed different patterns over time in the different risk groups (Fig 5). SF-6D utility score fluctuations over time were most pronounced among 18–59 year-olds, regardless of the presence of a medical risk condition. 60–85 year-olds showed more steady patterns of SF-6D utility scores over time, with patterns decreasing before stabilizing (those with a medical risk condition) or slightly increasing (those without a medical risk condition) from April 2022 onwards. For MCS, 18–59 year-olds without a medical risk condition showed strongest fluctuations over time, while 60–85 year-olds and 18–59 year-olds with a medical risk condition showed more steady patterns. Among 60–85 year-olds MCS patterns slowly decreased up until February 2022 and slightly increased thereafter. 18–59 year-olds with a medical risk condition showed an overall declining pattern up until November 2021, whereafter an increasing trend was observed. Lastly, PCS fluctuations over time were comparable between age groups, with 60–85 year-olds showing more steady patterns than those 18–59 years old. After a small peak in the second half of December 2021, decreasing patterns of PCS were observed in all groups between January and April 2022.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5.

Weekly smoothed mean of (A) health utility (SF-6D), (B) mental health (MCS), and (C) physical health (PCS) by risk group in the VASCO cohort between 17-05-2021 and 31-07-2022. * Even though SF-6D and MCS/PCS scores can respectively range from 0 to 1 (population average 0.8) and 0 to 100 (population average 50), y-axes are shortened (respectively 0.7–0.9 and 40–60) to better visualize differences.

https://doi.org/10.1371/journal.pone.0300324.g005

Statistically significant positive correlations between SF-6D utility scores and OxCGRT’s Stringency Index were found among those 18–59 years old without a medical risk condition (p = 0.001) and 60–85 year-olds with (p = 0.018) and without a medical risk condition (p<0.001), but correlation coefficients were small (0.16, 0.12 and 0.30, respectively) (S1 File, Fig S1.8). For MCS scores, statistically significant positive correlations with the Stringency Index were observed for those without a medical risk condition (18–59 years: p = 0.026; 60–85 years: p = 0.006), but correlation coefficients were again small (18–59 years: 0.11; 60–85 years: 0.14) (S1 File, Fig S1.9). PCS scores were statistically significantly, positively correlated with the Stringency Index among those 18–59 years old without a medical risk condition and 60–85 year-olds with and without a medical risk condition (p<0.001 in all groups) (S1 File, Fig S1.10). Correlation coefficients for PCS scores were small to moderate (respectively 0.40, 0.21 and 0.41 for each risk group).

In our sensitivity analysis with varying lag times of zero, one, three and four weeks no large differences were observed in the correlations between the Stringency Index and our outcomes (S1 File, Table S1.4). For SF-6D utility scores, statistically significant correlations were found for the same risk groups in all four lag time variations and correlation coefficients were comparable to those found in the two-week lag time analysis. Observed correlation coefficients for MCS scores were also comparable to those of the two-week lag time analysis. However, contrary to the two-week lag time analysis, MCS scores of those 18–59 years old without a medical risk condition were not statistically significantly correlated with the Stringency Index in the three- and four-week lag time analyses (p = 0.115 and p = 0.253 respectively). Additionally, in the four-week lag time analysis, 60–85 year-olds with a medical risk condition did show a statistically significant positive correlation between MCS scores and the Stringency Index (p = 0.014). For PCS scores, varying lag times also did not greatly influence correlation coefficients of the correlation with the Stringency Index. However, contrary to the two-week lag time analysis, a statistically significant positive correlation between PCS scores and the Stringency Index was observed among those 18–59 years old with a medical risk condition in the one-week lag time analysis (p = 0.016).

The sensitivity analysis comparing the pre-Omicron period and the Omicron period showed mostly non-significant correlations between the Stringency Index and our outcome measures, five statistically significant positive correlations and two negative ones: for MCS scores of 60–85 year-olds with (correlation coefficient: -0.24; p = 0.002) and without a medical risk condition (correlation coefficient: -0.25; p<0.001) during the Omicron period (S1 File, Table S1.5).