A Non-Randomised Controlled Study of Interventions Embedded in the Curriculum to Improve Student Wellbeing at University

Abstract

Universal and preventative interventions are advocated via the curriculum and pedagogy to help overcome the increasing prevalence of poor mental health among university students. To date, the literature in this field is overall of poor quality and cannot be synthesised for meta-analysis, due to poor reporting of methodology and results, lack of control conditions, and mixed outcomes across studies. This study examines the effectiveness of curriculum-embedded interventions on student wellbeing at university. A non-randomised design compared four curriculum-embedded interventions with matched controls from the same cohort (Psychology, English, Nursing, International Politics). To increase power, a meta-analytic approach combined the conditions to examine improvements in student wellbeing, social connectedness, loneliness, students flourishing, self-compassion, burnout, self-esteem, and learning approach. There were non-significant improvements in the intervention versus control conditions across all outcomes. There is no strong support for curriculum-embedded interventions improving student wellbeing at university. Despite improvements in study design and reporting, the sample size was still a challenge. More studies of high quality need to be conducted to provide evidence to guide teaching staff in supporting student wellbeing in the curriculum. Qualitative research is required to fully understand students’ experiences.

1. Introduction

In the context of increasing prevalence of youth mental health problems [1,2,3], concern about university student mental health is mounting. A high prevalence of mental distress has been observed among students internationally [3,4,5]. There are increasing numbers of students reporting mental health conditions and seeking support through university counselling services worldwide, including the UK [6,7], USA [8], and Australia [9]. While the challenge of poor student mental health has often been discussed in the context of mental health services [10,11], recent guidance advocates for a ‘whole-university’ approach, whereby all aspects of university life should be targeted to promote positive wellbeing [12]. This approach is a settings-based model, with foundations in the World Health Organisation’s Ottawa charter, which states the following: “Our societies are complex and interrelated… Health is created and lived by people within the settings of their everyday life: where they learn, work, play and love” [13]. The settings-based approach to health views effective health promotion as a process of enabling all sectors of a community to engage creatively in driving toward better health [14]. The Okanagan charter for promoting health in universities and colleges elaborates that health promotion should not just be the responsibility of the health sector, but it must engage all sectors of the community [15].

Building on the settings-based approach, learning is one of the four themes addressed by the University Mental Health Charter [12]. Pedagogy and curriculum are fundamental to a settings-based approach to wellbeing, as these are the only guaranteed points of contact between a university and its students [6]. Strategies to support university student wellbeing in the curriculum can be direct, by embedding wellbeing content and teaching students directly about wellbeing and mental health, for example, through the ‘5 ways to wellbeing’ approach, and curriculum infusion [16]. Alternatively, indirect strategies, involving changes to pedagogy, curriculum and assessment, can be adopted to support wellbeing. This could include promoting peer connections through classroom activities, scaffolding student autonomy [16], redeveloping the curriculum to reduce stress (e.g., type of assessment, time of assessment), embedding inclusive pedagogy and assessment [17], supporting study skills [18], and fostering a psychological safe classroom environment [19].

While the University Mental Health Charter calls for universities to ensure that curriculum design, pedagogic practice and academic processes have a positive impact on the mental health and wellbeing of all students [12], evidence outlining how to achieve this is scarce [20]. A systematic review of settings-based interventions for mental health in universities identified 19 papers, predominantly focused on academic-based strategies, curriculum infusion and assessment strategies [21]. However, the review concluded that the evidence related to the effectiveness of such interventions was inconclusive and noted that the internal validity of the data was very low, with few studies incorporating a control group [21]. A more recent review of the literature from 2015 to 2020 found that despite the increase in studies (N = 46) in this field, due to poor and inconsistent reporting across studies, there was still no strong evidence supporting the effectiveness of curriculum-embedded interventions for improving student wellbeing at university [22]. Many of the studies were underpowered, highlighting the challenge of achieving a powered sample size when studying specific cohorts of university students. A solution would be to employ a meta-analytic approach to combine studies and increase power. However, meta-analyses can only be achieved if raw data are available to calculate effect sizes, a clear description of the intervention and control conditions is provided, and there are consistent outcome measures across interventions. This was not possible with the current literature to date.

To effectively implement a settings-based approach, universities and academics require guidelines that outline the process by which curriculum design, pedagogy practice and academic processes can positively impact student mental health and wellbeing. This sector requires further evidence, evaluating different possible strategies, for these guidelines to be developed. To address this gap, we have evaluated a range of curriculum-embedded interventions across one large university, designed to promote positive wellbeing. We measured the same outcomes across interventions. In the existing literature, researchers have utilised many different types of wellbeing measures. To develop a questionnaire for this study, we reviewed outcomes assessed in systematic reviews in this field [21,22]. In addition, we had conversations with module leaders of the interventions to understand which elements of wellbeing they aimed to improve. This resulted in the following list of validated outcome measures for this research: social connectedness (Social Connectedness Scale—SCS) [23], loneliness (UCLA-6) [24], mental wellbeing (Short-Form Warwick Edinburgh Mental Well-Being Scale—SWEMWBS) [25], flourishing (Flourishing Scale) [26], self-compassion (Self-Compassion Short-Form Scale—SC-SF) [27], burnout (Short Burnout Scale) [28] and self-esteem (Single-Item Self-Esteem Scale—SISES) [29].

Objectives

To evaluate the effectiveness of curriculum-embedded interventions on improving wellbeing outcomes of undergraduate students.

2. Methods

This study informed the ‘Education for Mental Health’ project [19], a national online toolkit for university teaching staff that outlines strategies to support student wellbeing through the curriculum.

This study employed a non-randomised controlled 2 × 2 mixed study design, with two timepoints (pre- and post-module) comparing modules that aim to improve student wellbeing (intervention) to modules that did not aim to improve wellbeing (control). To improve power, a meta-analytic approach combined the intervention and control groups. Ethical approval was obtained (ref. LRS-19/20-15013). After reviewing the Enhancing the QUAlity and Transparency Of health Research (EQUATOR) library, the ‘Guidelines for Reporting Non-Randomised Studies’ were selected to report the methods of this study [30].

2.1. Participants and Setting

Participants included undergraduate students who were enrolled in the intervention or control modules outlined below. The only exclusion criteria included not being enrolled on the intervention or control modules.

A convenience sample of students was recruited using a range of methods, including virtual learning environments, emails from module leaders, small presentations from the research team within lectures or tutorials, university research participation systems, and social media. Recruitment adverts signposted students to an online survey hosted on Qualtrics containing an information sheet, consent form, and asking students to leave their email address. From here, Qualtrics created an ID number, and automatically sent students the Time 1 survey and, at the appropriate time point, the Time 2 survey. This method was used to minimise the attrition rate that can be high when asking survey respondents to create their own ID number to match multiple timepoints [31]. The ID number attached to survey responses allowed the surveys to remain anonymous when exporting data from Qualtrics. Students could choose to enter a prize draw to win an iPad upon completion of both surveys.

Students were asked to self-report demographic information, as recorded in

Table 1. Pre-module student demographic data for intervention and control modules by subject.

	Psychology (All Years)			English Studies (All Years)			Nursing (1st Years)			International Politics (2nd Years)
	Intervention N = 50	Control N = 25	p-Value	Intervention N = 11	Control N = 6	p-Value	Intervention N = 12	Control N = 8	p-Value	Intervention N = 24	Control N = 17	p-Value
Year of study, N(%)			0.06			0.002 *			-			-
1st year	15 (30)	2 (8)		9 (82)	0 (0)		12 (100)	8 (100)		-	-
2nd year	24 (48)	13 (52)		0 (0)	3 (50)		-	-		24 (100)	16 (100)
3rd year	11 (22)	10 (40)		2 (18)	3 (50)		-	-		-	-
Age (years), N(%)			0.02 *			1.00			0.51			1.00
17–20	41 (84)	14 (56)		9 (90)	5 (83)		9 (75)	4 (57)		16 (67)	11 (65)
21–24	6 (12)	8 (32)		1 (10)	1 (17)		1 (8)	0 (0)		7 (29)	6 (35)
25+	2 (4)	3 (12)		0 (0)	0 (0)		2 (17)	3 (43)		1 (4)	0 (0)
Gender, N(%)			0.26			1.00			0.61			0.03*
Female	48 (98)	23 (92)		8 (80)	5 (83)		11 (92)	6 (86)		16 (67)	5 (29)
Male	1 (2)	2 (8)		2 (20)	1 (17)		1 (8)	0 (0)		8 (33)	12 (71)
I use another term	0 (0)	0 (0)		0 (0)	0 (0)		0 (0)	1 (14)		0 (0)	0 (0.0)
Ethnicity, N(%)			0.94			0.79			0.16			0.32
White British	10 (20)	5 (20)		4 (40)	2 (33)		3 (25)	2 (29)		8 (33)	3 (18)
Other white background	12 (25)	5 (20)		3 (30)	1 (17)		0 (0)	2 (29)		7 (29)	9 (53)
BAME $	27 (55)	15 (60)		3 (30)	2 (33)		9 (75)	3 (43)		9 (38)	5 (29)
Prefer not to say	0 (0.0)	0 (0)		0 (0)	1 (17)		0 (0)	0 (0)		0 (0)	0 (0)
Part-time job, N(%)			0.47			0.22			0.07			0.01*
Yes	11 (22)	7 (28)		6 (60)	1 (17)		6 (50)	0 (0)		12 (50)	3 (18)
No	38 (78)	18 (72)		4 (40)	5 (83)		6 (50)	7 (100)		12 (50)	14 (82)
Student status, N(%)			0.20			0.46			0.37			0.52
Home (UK)	24 (49)	18 (72)		9 (90)	4 (68)		12 (100)	6 (86)		12 (50)	5 (29)
EU	10 (20)	3 (12)		1 (10)	1 (17)		0 (0)	1 (14)		6 (25)	6 (35)
International	15 (31)	4 (16)		0 (0)	1 (17)		0 (0)	0 (0)		6 (25)	6 (35)
Disability status, N(%)			0.17			1.00			0.52			1.00
Yes	2 (4)	21 (84)		2 (20)	1 (17)		1 (8)	2 (29)		1 (4)	1 (6)
No	46 (94)	4 (16)		8 (80)	5 (83)		11 (92)	5 (71)		22 (92)	15 (88)
Prefer not to say	1 (2)	0 (0)		0 (0)	0 (0)		0 (0)	0 (0)		1 (4)	1 (6)
First in family to study at university, N(%)			0.17			0.68			0.63			0.78
Yes	16 (33)	4 (16)		2 (20)	0 (0)		6 (50)	2 (29)		3 (13)	1 (6)
No	33 (67)	21 (84)		7 (70)	6 (100)		6 (50)	5 (71)		20 (83)	16 (94)
Unsure	0 (0)	0 (0)		1 (10)	0 (0)		0 (0)	0 (0)		1 (4)	0 (0)
Caring responsibilities, N(%)			1.00			1.00			0.31			0.29
Yes	3 (6)	2 (8)		1 (10)	0 (0)		2 (17)	3 (43)		1 (4)	3 (18)
No	46 (94)	23 (92)		9 (90)	6 (100)		10 (83)	4 (57)		23 (96)	14 (82)

^$ Including white and black Caribbean (n = 1), white and black African (n = 3), white and Asian (n = 6), any other mixed background (n = 2), Indian (n = 2), Pakistani (n = 2), Bangladeshi (n = 7), Chinese (n = 9), any other Asian background (n = 6), African (n = 2), Caribbean (n = 2); other ethnic group (n = 2). * Statistically significant p < 0.05.

, and had the option of selecting ‘prefer not to say’ to these demographic questions. Demographics information was obtained categorically to protect the anonymity of students. Individual-level Time 1 (pre-module) categorical data (year of study, age, gender, ethnicity, part-time job status, student status, disability status, first in family to study at university, caring responsibilities) were tabulated (N,%) and p-values were presented (Fischer’s exact test) to inspect the differences between intervention and control groups, as shown in Table 1.

2.2. Sample Size

The effect size in the Penberthy et al. [32] study for differences between the intervention and control groups was 0.51 (standardised mean difference—SDM) in the Self-Compassion Scale. Taking this effect size, alpha of 0.05 and power of 90%, the required sample size was 134 (minimum of 67 per group).

2.3. Interventions

The intervention and control conditions are described according to the TIDieR guidelines [33], adapted for this context, as shown in

Table 2. Summary of curriculum-embedded interventions according to TIDieR checklist.

Name	Number of Students Taking Module; Year of Study	Why: Wellbeing Approach	What	Who Provided	How/Where	When and How Much
Intervention modules
Psychology: Graduate Attributes	N = 296; All years	Academic-based strategy	Sessions on core study skills, digital literacy, skills for mental wellbeing, communication skills, and building a career.	Senior lecturers and teaching fellows in Psychology department	Online; mixture of synchronous and asynchronous (LinkedIn learning videos)	Two semesters (October 2020–March 2021); 20 sessions
English Studies: Skills and Support for your Degree	N = 220; All years	Academic-based strategy	Skills development for degree, career support to improve wellbeing.	Senior lecturers in English department	Online; synchronous	Two semesters (October 2020–March 2021); 22 sessions
Nursing: Wellbeing in London	N = 36; 1st years	Five ways to wellbeing	Supports students to engage in the ‘5 ways to wellbeing’ approach: keep moving, invest in relationships, never stop learning, give to others and savour the moment.	Lecturer in Nursing Education	Online; synchronous	One semester (January–March 2021); 6 sessions
International Politics: Issues in International Politics	N = 89; 2nd years	Curriculum infusion	Teaches theoretical and empirical content regarding international relations. Sessions are structured around human emotions.	Teaching fellow and graduate teaching assistant	Online; synchronous	One semester (January–March 2021); 10 sessions
Control modules
Psychology: Graduate Attributes	N = 348; All years	Less intensive intervention, i.e., same as intervention but attended <4 sessions				Two semesters (October 2020–March 2021); <4 sessions
English Studies: Skills and Support for your Degree	N = 380; All years	Less intensive intervention, i.e., same as intervention but attended <4 sessions				Two semesters (October 2020–March 2021); <4 sessions
Nursing: Mental Health in Context	N = 38; 1st years	Did not aim to improve student wellbeing	Teaches students about the multiplicity of mental health presentations, their different management styles, and the ethical issues that surround them.	Lecturer in Nursing Education	Online; synchronous	One semester (January–March 2021); 6 sessions
International Politics: Economics of the Public Sector	N = 46; 2nd years	Did not aim to improve student wellbeing	Teaches economic analysis of taxation and spending on the welfare state in the UK.	Reader in Political Economy	Online; synchronous	One semester (January–March 2021); 10 sessions

. Fidelity assessment was expressed via reporting of the level of teaching expertise in Table 2 under ‘Who Provided’.

Interventions were not developed explicitly for this study. Instead, academics were recruited via staff forums across one large university. Teaching staff or module leaders could approach the research team if they felt their module met the criterion of “aiming to improve student mental wellbeing” and were interested in their student cohort taking part in the research. Modules could target any undergraduate student in any year or subject and could be delivered via any mode (face-to-face or online), be compulsory or optional, credited or non-credited. Undergraduate modules within the following subjects were identified as the ‘intervention’ conditions: Psychology, Nursing, English studies, and International Politics. All interventions identified were optional modules and sessions were delivered weekly. The Nursing and Politics intervention and control modules carried credit and the Psychology and English studies modules were non-credit bearing.

The Psychology module focused on developing attributes students require after graduating, in order to be ready for work or further study, including communication skills, digital skills, and self-care. This used a combination of approaches, including stress management, mindfulness, behavioural self-care, and peer-led sessions. Likewise, the English studies module adopted an indirect approach to improving student wellbeing via academic-skills development. Both the Psychology and English studies modules targeted students across all undergraduate years, thus creating an opportunity to support students transitioning into university. This aligns with some of Sally Kift’s work on transition pedagogy and using the curriculum to help students adjust academically and socially to university [34]. To embed mental wellbeing in the curriculum, Houghton and Anderson [16] advocate utilising the ‘5 ways to wellbeing’ approach, and curriculum infusion, i.e., discipline-relevant mental health, wellbeing content, and resources [16]. These approaches were employed by the Nursing and International Politics modules, respectively. As well as incorporating techniques from the existing systematic review literature, the Nursing module structured sessions around the ‘5 ways to wellbeing’ approach, which encourages students to keep moving, invest in relationships, never stop learning (art-based session), give to others (peer-led session), and savour the moment (mindfulness techniques). The International Politics module uses a curriculum infusion approach by teaching theoretical and empirical content regarding international relations, structured around different human emotions, e.g., “Anger: Is violence political?”. In addition, the International Politics module leader aimed to embed wellbeing within pedagogy, i.e., employing strategies to make learning a more collaborative process, increase autonomy and reduce student loneliness.

Control conditions were identified with the help of module leaders. As the Psychology and English studies modules targeted all year groups and were open to the whole cohort, the control condition for these were students who attended fewer than four sessions of the intervention. For the Nursing and International Politics cohort, the control conditions included students from same subject and year who were enrolled on a module that was mutually exclusive from the intervention condition. These modules were matched in terms of duration and frequency, mode of delivery, and delivered within the same department, by staff in the same department. These control conditions did not aim to improve student wellbeing.

2.4. Outcomes

Measures of social connectedness (SCS [23]), loneliness (UCLA-6 [24]), mental wellbeing (SWEMWBS [25]), flourishing (Flourishing Scale [26]), self-compassion (SC-SF [27]), burnout (Short Burnout Scale [28]) and self-esteem (SISES [29]); were included as measures of student wellbeing. The Revised Study Process Questionnaire (R-SPQ-2F [35]) was used to assess learning approaches, i.e., deep or surface learning. Students were also asked to rate their agreement (5-point scale from strongly disagree from strongly agree) to the following statement: ‘Universities should embed a wellbeing focus within the curriculum’.

Table 3. Online survey items from Timepoint 1 (pre-module).

Measure	Number of Items	Scoring	Scores Range from…	What Scores Represent	Example Item	Internal Consistency (Cronbach’s Alpha)
SCS*	12	1–7 Strongly disagree to strongly agree	12 to 84	Higher scores = higher campus connectedness	There are people at university with whom I feel a close bond.	0.92 [36]
UCLA-6	6	1–4 Never, rarely, sometimes, often	6 to 24	Higher scores = greater loneliness	I lack companionship.	0.82 [24]
SWEMWBS	7	1–5 None of time, rarely, some of time, often, all of time	7 to 35	Higher scores = more positive mental wellbeing	I have been feeling optimistic about the future.	0.84 [37]
Flourishing Scale *	8	1–7 Strongly disagree to strongly agree	8 to 56	Higher scores = greater sense of flourishing	I lead a purposeful and meaningful life.	0.89 [38]
SC-SF	12	1–5 Almost never to almost always	12 to 60	Higher scores = higher self-compassion	I try to be understanding and patient towards those aspects of my personality I do not like.	0.86 [27]
Short Burnout	3	1–6 Strongly disagree to strongly agree	3 to 18	Higher scores = higher burnout	I feel burned out from my studies.	Exhaustion and cynicism subscales = 0.66 and 0.79, respectively [28]
SISE	1	1–5 Strongly disagree to strongly agree	1 to 5	Higher scores = higher self-esteem	I have high self-esteem.	0.75 [29]
R-SPQ-2F: Deep learning approach subscale	10	1–5 Never or only rarely true of me, sometimes true of me, true of me about half the time, frequently true of me, always or almost always true of me	10 to 50	Higher scores = greater use of deep learning approach	I find that, at times, studying gives me a feeling of deep personal satisfaction.	0.73 [35]
R-SPQ-2F: Surface learning approach subscale	10	1–5 Never or only rarely true of me, sometimes true of me, true of me about half the time, frequently true of me, always or almost always true of me	10 to 50	Higher scores = greater use of surface learning approach	My aim is to pass the course while doing as little work as possible.	0.64 [35]

* These measures were dropped at Time 2, see ‘Factor analysis to determine wellbeing measures’ section for details.

describes the survey measures in full.

2.5. Factor Analysis to Determine Wellbeing Measures

Upon inspection of Time 1 survey (pre-module) data, researchers identified potential overlap between certain wellbeing measures. In addition, feedback from students who completed the Time 1 survey revealed that the survey took too long to complete. Therefore, a factor analysis was run to examine the overlap between the measures and reduce the overall survey length for Time 2 (post-module) and maximise retention.

Time 1 data were available for 192 students. Principal component analysis (PCA) was run on 45-items, including the following measures: SCS, UCLA-6, SWEMWBS, Flourishing Scale, SC-SF. The short burnout and SISE scales were not included as they were already shortened measures, i.e., three items and one item respectively. Inspection of the correlation matrix showed that all variables had at least one correlation coefficient greater than 0.3. The overall Kaiser–Meyer–Olkin (KMO) measure was 0.92, with individual KMO measures all greater than 0.8, this was classified as ‘meritorious’ to ‘marvellous’ according to [39]. Bartlett’s test of sphericity was statistically significant (p < 0.001), indicating that the data were to be likely factorisable.

PCA revealed only three components that each explained at least 5% of the total variance, explaining 31.59%, 10.27%, and 5.79% of the total variance, respectively. Visual inspection of the scree plot indicated that three components should be retained [40]. In addition, a three-component solution met the interpretability criterion. As such, three components were retained.

The three-component solution explained 47.66% of the total variance. Varimax orthogonal rotation was employed to aid interpretability. The rotated solution exhibited a ‘simple structure’ [41]. Component loadings and communalities of the rotated solution are presented in Table S1. Component 1 shows the overlap between the campus connectedness and UCLA-6 measures. Component 2 shows overlap between the SWEMWBS and Flourishing Scale. All items of the SC-SF load onto component 3.

There was a statistically significant, strong negative correlation between total campus connectedness and total UCLA-6 scores, r_s(189) = −0.72, p < 0.001. There was a statistically significant, strong positive correlation between total metric SWEMWBS and total Flourishing Scale scores, r_s(189) = −0.77, p < 0.001. Due to the SWEMWBS and UCLA-6 being more widely implemented than the campus connectedness and Flourishing Scales, the latter scales were dropped from the Time 2 survey.

2.6. Time 2 Survey Measures

The Time 2 survey included the following measures, as described above: SWEMWBS, UCLA-6, SC-SF, SISES, Short Burnout Scale, and R-SPQ-2F.

2.7. Blinding

The researcher (RU) that assessed the outcomes was not blinded to the participants’ exposure, i.e., intervention or control, but participants responses were non-identifiable.

2.8. Statistical Methods

An aggregate meta-analysis combined data from the intervention and control conditions across modules, for each wellbeing outcome (SWEMWBS, UCLA-6, SC-SF, SISES, Short Burnout Scale) and the R-SPQ-2F. The SMD, Cohen’s d, was calculated in STATA 14 [42] to determine the change in outcomes between Time 1 and Time 2 surveys (pre-post module). A random effects meta-analysis was used to pool SMDs. Forest plots provided a visual representation of the meta-analyses. This method allowed analysis of a pooled effect across all modules, and to inspect each module separately.

3. Results

3.1. Participant Flow

As summarised in Figure 1, there were a total of 1453 students who enrolled in the 4 intervention and control modules and 153 students completed Time 1 surveys; therefore, there was a 10.53% response rate. As 138 completed Time 2 surveys, only 15 students (9.80%) were lost to follow-up.

3.2. Baseline Data

Table 1 displays Time 1 (pre-module) demographic data for the intervention and control groups by subject. For Psychology students, there was a significant association between condition (intervention vs control) and age, with a higher percentage of students in the intervention module aged 17–20 (83.7%) than the control module (56.0%). For English studies students, there was a significant association between conditions and year of study, with a higher percentage of students in the 1st year (81.8%) than the control condition (0.0%). There were no significant associations between intervention and control conditions and demographics for Nursing students. For International Politics students, there was a significant association between conditions and gender, with a higher percentage of females in the intervention module (66.7%) than the control module (29.4%). Furthermore, there was a significant association between conditions and employment status with a higher percentage of students in the intervention module with a part-time job (50.0%) than the control module (17.6%).

3.3. Outcomes and Estimation

Raw means and standard deviations at Time 1 and Time 2 (within mean differences) for the intervention and control groups, by subject, are presented in

Table 4. Outcome data pre and post module for intervention and control conditions by subject.

		Psychology				English Studies				Nursing				International Politics
Outcome		Intervention		Control		Intervention		Control		Intervention		Control		Intervention		Control
		N	M(SD)	N	M(SD)	N	M(SD)	N	M(SD)	N	M(SD)	N	M(SD)	N	M(SD)	N	M(SD)
Wellbeing	Pre	50	21.40 (3.93)	25	20.36 (3.41)	11	19.22 (2.20)	6	19.31 (2.74)	8	19.94 (2.96)	6	20.27 (2.91)	19	19.19 (4.37)	13	21.95 (4.60)
	Post	50	20.51 (3.87)	25	19.22 (3.35)	11	18.53 (4.62)	6	17.67 (5.83)	8	21.59 (3.42)	6	21.83 (2.80)	19	19.79 (3.06)	13	21.17 (3.96)
	Mean difference	−0.89		−1.14		−0.69		−1.64		1.65		1.56		−0.60		−0.78
Loneliness	Pre	50	14.46 (4.40)	25	15.00 (5.23)	11	17.64 (3.93)	6	15.33 (4.46)	8	16.25 (3.85)	6	18.33 (3.20)	19	14.42 (3.83)	13	14.46 (4.47)
	Post	50	14.64 (4.26)	25	16.60 (5.37)	11	16.55 (5.05)	6	16.83 (3.97)	8	15.38 (3.58)	6	14.67 (2.07)	19	13.79 (3.90)	13	13.54 (5.03)
	Mean difference	0.18		1.60		−1.09		1.50		−0.87		−3.66		−0.63		−0.92
Self-compassion	Pre	50	35.22 (8.54)	25	33.16 (8.84)	11	28.82 (6.00)	6	30.83 (2.86)	8	30.88 (9.98)	6	35.50 (10.73)	18	36.06 (7.82)	13	38.23 (8.79)
	Post	50	35.14 (8.24)	25	33.20 (6.65)	11	30.81 (8.36)	6	27.33 (4.23)	8	34.00 (8.82)	6	37.67 (11.13)	18	35.06 (8.34)	13	35.23 (7.01)
	Mean difference	−0.08		0.04		1.99		−3.50		3.12		2.17		−1.00		−3.00
Self-esteem	Pre	49	2.76 (1.11)	25	2.48 (1.05)	11	2.20 (0.79)	6	3.17 (0.98)	8	3.00 (0.93)	6	2.33 (0.82)	19	3.47 (1.17)	13	3.77 (0.83)
	Post	49	3.04 (1.11)	25	2.60 (1.04)	11	2.00 (0.78)	6	1.83 (0.98)	8	3.13 (1.13)	6	2.33 (0.82)	19	3.53 (1.22)	13	3.85 (0.80)
	Mean difference	0.28		0.12		−2.00		−1.34		0.13		0		0.06		0.83
Burnout	Pre	50	9.04 (3.72)	25	10.96 (4.07)	11	11.10 (3.21)	6	11.33 (5.20)	7	11.86 (3.98)	6	9.67 (4.03)	19	11.68 (4.45)	13	11.62 (2.69)
	Post	50	11.76 (4.17)	25	14.64 (3.09)	11	12.64 (3.04)	6	13.50 (4.59)	7	10.00 (2.24)	6	9.50 (3.83)	19	13.84 (4.23)	13	12.23 (4.87)
	Mean difference	2.72		3.68		1.54		2.17		−1.86		−0.17		2.16		0.61
Deep learning approach	Pre	50	32.36 (6.22)	25	28.76 (6.88)	11	31.10 (6.87)	6	28.17 (10.07)	7	27.43 (9.05)	6	35.83 (3.66)	18	29.61 (7.36)	13	33.23 (6.44)
	Post	50	29.44 (6.82)	25	26.60 (6.80)	11	30.18 (9.13)	6	25.50 (12.28)	7	31.86 (8.60)	6	34.33 (3.61)	18	29.00 (9.27)	13	32.08 (7.06)
	Mean difference	−2.92		−2.16		−0.92		−2.67		4.43		−1.5		−0.61		−1.15
Surface learning approach	Pre	50	23.52 (7.04)	25	23.80 (8.13)	11	21.70 (4.69)	6	23.83 (7.81)	7	26.57 (6.80)	6	18.17 (4.83)	18	26.67 (8.73)	13	24.92 (4.72)
	Post	49	24.33 (7.70)	25	25.32 (8.15)	11	20.18 (4.85)	6	28.33 (11.72)	7	27.71 (7.39)	6	20.00 (6.99)	18	26.89 (8.13)	13	25.54 (7.48)
	Mean difference	0.81		1.52		−1.52		4.50		1.14		1.83		0.22		0.62
Should universities embed wellbeing in the curriculum?	Pre	50	4.12 (0.69)	25	3.92 (0.86)	11	4.18 (1.17)	6	4.17 (1.17)	8	3.75 (1.17)	6	4.33 (0.52)	19	4.05 (0.91)	13	3.85 (1.46)
	Post	50	4.04 (1.11)	25	4.08 (0.86)	11	3.91 (1.14)	6	3.83 (1.17)	8	3.88 (1.36)	6	3.83 (1.47)	19	4.21 (1.03)	13	4.00 (1.08)
	Mean difference	−0.08		0.16		−0.27		−0.33		0.13		−0.50		0.16		0.15

M = mean; N = sample size; SD = standard deviation.

. As displayed in Figures S1–S7, for all outcomes, heterogeneity was low and non-significant (p > 0.05).

3.4. Wellbeing Outcomes

In a series of random effects meta-analyses, there were non-statistically significant improvements for the intervention groups compared with the control groups in mental wellbeing (SMD = 0.10, 95% CI = −0.25 to 0.44), as shown in Figure S1, loneliness (SMD = −0.10, 95% CI = −0.56 to 0.36), as shown in Figure S2, self-compassion (SMD = 0.20, 95% CI = −0.19 to 0.58), Figure S3, and burnout (SMD = −0.10, 95% CI = −0.45 to 0.25), as shown in Figure S4.

There was no change in self-esteem for the intervention groups compared with the control groups (SMD = 0.00, 95% CI = −0.35 to 0.35), as shown in Figure S5.

3.5. Learning Approach

In a random effects meta-analysis, there was a non-statistically significant increase in deep learning for the intervention groups compared with the control groups (SMD = 0.06, 95% CI = −0.29 to 0.41), as shown in Figure S6. There was a non-statistically significant decrease in surface learning for the intervention groups compared with the control groups (SMD = −0.19, 95% CI = −0.54 to 0.16), as shown in Figure S7.

3.6. Views on Embedding Wellbeing in the Curriculum

Students rated their agreement to the following statement: ‘Universities should embed a wellbeing focus within the curriculum’ at Time 1 and 2. In a random effects meta-analysis, there was a non-statistically significant decrease in agreement to universities embedding wellbeing in the curriculum for the intervention groups compared with the control groups (SMD = −0.09, 95% CI = −0.44 to 0.26).

4. Discussion

This study aimed to understand whether settings-based approaches within the university curriculum can improve student wellbeing. The efficacy of four curriculum-based wellbeing interventions versus control modules from Psychology, English studies, Nursing and International Politics cohorts were investigated. The findings failed to demonstrate a significant impact of the intervention modules on student wellbeing and learning approach outcomes compared to control modules. This is consistent with the overall findings from two systematic reviews of settings-based approaches [21,22]. The modules evaluated drew on approaches outlined in individual studies synthesised in the systematic reviews, e.g., stress-management, mindfulness, behavioural self-care, and peer-led sessions. However, they were unique in presentation, for example, using skills-development strategies, the ‘5 ways to wellbeing’ approach and curriculum infusion.

The students did not significantly change their view from pre- to post-intervention on whether they think universities should embed a wellbeing focus within the curriculum, with, on average, students agreeing to this suggestion. There were no significant differences between the intervention and control conditions, indicating that, overall, students were happy with settings-based approaches to improving wellbeing within the curriculum. This should be further supported with qualitative research and student co-creation groups to examine preferences and impact on wellbeing [19,43].

Albeit statistically non-significant, improvements across wellbeing (except self-esteem, no change) and learning approaches outcomes across the intervention conditions should be noted. Despite meeting the sample size determined by the power calculation, the overall sample was still relatively small (N = 138). A meta-analytic approach to combine settings-based approaches across student cohorts to improve power was used; however, the effects of the interventions at the individual module level were still very small. To observe real changes, it may be necessary to think about implementing these changes at a programme level, i.e., reaching across multiple modules. Interventions at the module level, rather than programme level, may be doomed to fail, because they require a small change across a much wider programme. If the rest of the programme is having a negative impact on the student’s mental health, it is unlikely that one module can make much difference. While it is argued that every area of work in this space is valuable, the null results around individual modules should not be discarded; they may rather be an indication that change needs to be greater and broader than individual modules. Further qualitative research in this space is recommended, as it may explore the poor effectiveness of the interventions.

4.1. Strengths

Overall, this study addresses the flaws of previous research synthesised in systematic reviews in this field [21,22], for example, by improving reporting quality by the use of guidelines [30,33], providing a detailed description of intervention and control conditions, thorough account of methodology, and reporting of raw statistics (Table 4) and effect sizes (SMDs). In addition, consistent wellbeing outcomes were measured across the interventions. This is a common complexity in this field, for instance, a scoping review found 28 different validated measures of wellbeing [44]. A factor analysis to only utilise the necessary wellbeing outcomes was adopted. Previous studies failed to include control groups or use cross-sectional and time-series designs, which left their results prone to confounding [21,22]. In the present research, all intervention conditions had a control condition to minimise bias. The attrition rate was relatively low (9%), highlighting the strength of the recruitment methods.

4.2. Limitations

Despite carefully developing diverse strategies of communication to recruit students through consulting with module leaders and student representatives from each subject cohort, the response rate was still low (10%). This was exaggerated by two of the interventions involving whole cohorts, in addition to being non-credited (Psychology and English studies), and by the response rate being higher for the Nursing and International Politics modules that were smaller in size and credited. Future research should aim to understand the recruitment challenges better and continue to involve students in the research process to maximise the engagement of other students.

Although the use of a real-world setting increases the ecological validity of the results, it did not allow for randomisation, as students self-select the modules to enrol in; the results may, therefore, be prone to selection bias. Using propensity score analysis would have increased the validity of the results [45].

It must be considered whether these findings demonstrate that modules designed to improve student wellbeing are not effective, or whether there are alternative explanations for these non-significant results. Although this study included a matched control group, group differences in gender and part-time job status were not controlled for in the analysis, due to sample size limitations. As both part-time employment and gender can impact student wellbeing, these group differences could have affected the results [46,47]. Additionally, it is possible that the intensity of exposure to the intervention varied significantly between students (e.g., due to differences in engagement, use of resources, attendance, etc.), which could also have affected the impact of the interventions on wellbeing. For the Nursing control condition, the module was not designed to improve student mental wellbeing and was not focused on personal wellbeing. However, the module content did refer to mental health conditions and exposure to this may have influenced student wellbeing. For the skills-development-based Psychology and English studies modules, lack of attendance did not necessarily equate to control participants not possessing the associated skills. Students may have obtained academic skills elsewhere, (e.g., from previous education); therefore, if skills acquisition equates to improved wellbeing, this might explain the lack of group differences. Future studies should aim to measure intervention exposure intensity and test for a dose–response relationship, as this would strengthen any causal inferences drawn [48].

Intervention reporting was strengthened via use of the TIDieR guidelines; however, only full reporting of the level of teaching expertise was conducted (teaching assistant, fellow, lecturer etc). Teaching observations would be optimal to evaluate whether interventions were being delivered as intended to a high standard. However, objective, non-biased observations and evaluation might not be feasible in most real-world teaching contexts. Future research, evaluating curriculum-embedded interventions, could tie in with module evaluation feedback, to examine how well students think teaching staff are delivering the interventions and whether they feel it impacts wellbeing.

Due to ethical considerations and protecting student anonymity, academic outcomes were not assessed. This could be important, as academic outcomes could have an impact upon exposure (intervention and control) and wellbeing outcomes [49,50].

4.3. Generalisability

Students self-selected the modules of interest; therefore, there is a chance that students only participated if they had an interest in wellbeing or personal mental health difficulties. However, overall, a diverse sample was achieved according to year of study, age, ethnicity, disability status, and caring status. This means findings are likely to be generalisable to other UK student cohorts.

For all modules evaluated, sessions were delivered online due to the COVID-19 pandemic. This could impact the generalisability to future modules conducting in-person sessions. However, some sessions would be delivered online anyway (e.g., the Psychology asynchronous sessions) and from discussions with module leaders, some sessions are likely to continue online or at least inform development of future sessions.

The real-world nature of this study meant the four interventions were heterogenous, for example, in terms of strategies for supporting wellbeing within the curriculum, module design (for example, length, duration, frequency, mode), and subject cohort. Therefore, even though they aim to achieve similar outcomes in terms of student wellbeing, their differences limit the generalisability of the findings.

5. Conclusions

Evaluating real-world, curriculum-embedded interventions to improve student wellbeing is a challenge. Small sample sizes make interpretation and generalisability questionable. Quantitative data need to be supported with qualitative research to fully understand how settings-based approaches within the curriculum impact wellbeing and suggestions for future development. Future research should consider programme level interventions to increase power or high-quality research that can be synthesised via meta-analyses.