Intentional Imperfection Program: A pilot randomised controlled trial to help university students manage perfectionism

Perfectionism, defined as having unrelenting standards for oneself and/or others, has increased in university students from 1989 to 2016 (Curran & Hill, 2019). Perfectionistic students encounter more interpersonal difficulties (Mackinnon et al., 2014), report lower academic adjustment and experience heightened psychological distress (Rice et al., 2006) compared to students who are less preoccupied with perfection. Although evidence-based treatments for perfectionism exist (Smith et al., 2023), few university students are likely to engage with these programs due to their lengthy format, typically involving 8–12 weekly/bi-weekly group or individual sessions. University students in need of psychological services often cite time and cost as barriers to seeking treatment (Abdel-Hady et al., 2022). Further, cognitive behavioural therapy for perfectionism may reduce perfectionistic self-related attitudes but appears to be ineffective in reducing traits found to be increasing in university students (Curran & Hill, 2019; Smith et al., 2023). Although perfectionistic attitudes and traits are linked, traits also encompass socially-based attitudes, cognitions, and behaviours (Sherry et al., 2003). Thus, there is a need for a brief, accessible, and efficacious intervention to help university students manage their perfectionistic traits.

The Perfectionism Social Disconnection Model (PSDM) describes how perfectionistic traits lead to psychological distress and disorder (Hewitt et al., 2017). The model posits that perfectionism can be socially prescribed (believing others have high standards for oneself), self-oriented (having high standards for oneself), and/or other oriented (having high standards for others; Hewitt et al., 2017). These traits can make individuals hypersensitive to interpersonal rejection and/or lead to off-putting interpersonal behaviours (e.g., coldness, aggression, aloofness) in social situations, resulting in social disconnection and psychological distress over time (Hewitt et al., 2017). Some longitudinal studies have indicated that perceived and actual social disconnection mediates the link between some perfectionistic traits and depressive symptoms (for meta-analysis, see Smith et al., 2020). In addition, some cross-sectional research has linked these perfectionistic traits with rejection sensitivity and interpersonal hostility (Magson et al., 2019). Thus, certain perfectionistic traits may lead to psychological distress through social disconnection, which could be targeted in treatment.

Newer research suggests that deficits in emotion regulation may also help to explain the mechanisms between perfectionism and psychological distress. Malivoire et al.’s (2019) conceptual model of emotion regulation in perfectionism suggests that perfectionistic individuals may experience negative affect in response to criticism/rejection, which may lead to unhelpful emotion regulation strategies and goals (e.g., favouring short-term relief despite negative consequences) when combined with low emotional awareness (Malivoire et al., 2019). Moreover, adaptive emotion regulation skills, such as self-compassion, mindfulness, and distress tolerance, may help perfectionistic individuals manage negative affect (Malivoire et al., 2019). In support, studies have found perfectionism to be positively associated with maladaptive emotion regulation strategies, such as avoidant coping (Castro et al., 2017). In addition, both self-compassion and mindfulness have been shown to moderate the association between perfectionism and psychological distress (Mehr & Adams, 2016; Tobin & Dunkley, 2021), and distress tolerance has been negatively associated with perfectionism (Rand-Giovannetti et al., 2022). As such, helping university students improve their emotion regulation skills may aid them in managing their perfectionism.

To explore this, we previously conducted an uncontrolled pilot study to test the preliminary feasibility and acceptability of a novel 2-hour online perfectionism program comprising techniques designed to improve university students’ emotion regulation (Visvalingam et al., 2022). Quantitative data showed statistically significant, small to moderate pre–post reductions in self-oriented perfectionism, socially prescribed perfectionism, hostility, rejection sensitivity, depression, and anxiety, as well as a small increase in perceived support. Thematic analyses of qualitative data indicated that participants found the program to be acceptable, enjoyable, and useful. Despite these promising results, other-oriented perfectionism remained unchanged, and participants requested more flexibility and additional time to complete the modules. Thus, the program was revised to place greater emphasis on addressing other-oriented perfectionism and offer increased completion flexibility.

In the current pilot trial, the primary objective was to examine the acceptability of the revised version of the online, un-guidedperfectionism program (i.e., Intentional Imperfection Program; IIP) and to determine the feasibility of a larger, randomised controlled trial. The secondary objective was to assess potential efficacy and obtain parameter estimates for our outcome measures to inform the design of a future definitive randomised controlled trial. Unlike most prior perfectionism trials that employed waitlist control comparators (Smith et al., 2023), we used an attention placebo control (i.e., Healthy Body and Mind Program; HBMP) to account for common therapeutic factors. Our primary efficacy outcome measure was psychological distress, while perfectionism, rejection sensitivity, interpersonal hostility, loneliness, academic resilience, self-compassion, mindfulness, and distress tolerance were secondary.

Trial design

We conducted a parallel-group, two-arm pilot randomised controlled trial, randomly allocating participants to either the IIP or the attention placebo control condition (i.e., HBMP). A randomisation ratio of 1:1 was used to ensure equal sample sizes across the arms of the pilot trial. No changes were made to the methods after the trial commenced, and this study was approved by the Macquarie University Human Research and Ethics Committee (no. 520231170148299).

Participants

First-year undergraduate psychology students (n = 1388) at Macquarie University provided consent and completed an online screening survey, which included the Hewitt and Flett Multidimensional Perfectionism Scale-Short Form (Hewitt et al., 2008). Students who reported high levels of perfectionism [i.e., scored one standard deviation greater than the mean of previous college samples for at least 1 subscale (self-oriented: > 27.96, other-oriented: > 22.87, socially prescribed: > 26.78; Stoeber, 2018)] and passed an attention check were invited to complete the baseline assessment to further confirm their eligibility (n = 577). One hundred twenty provided consent and completed the baseline assessment. To be included in the study, students needed to pass an attention check and also report high levels of perfectionism, but this time on the full version of the Hewitt and Flett Multidimensional Perfectionism Scale [i.e., scored one standard deviation greater than the mean of previous college samples for at least 1 subscale (self-oriented: > 79.28, other-oriented: > 65.93, socially prescribed: >61.05; Hewitt et al., 1991)] and on the concerns over mistakes subscale of the Frost Multidimensional Perfectionism Scale (Frost & Marten, 1990). A cut-off score of ≥ 25 on the concern over mistakes subscale was used, as this corresponds with clinically significant levels of perfectionism (Egan et al., 2014). Exclusion criteria were current engagement with treatment for perfectionism and/or initiating psychotropic medication within the past month. Figure 1 displays a CONSORT flow chart depicting participant flow in the study.

A CONSORT flowchart displaying participant flow through the study

Full size image

Interventions

A summary of the interventions can be found in Appendix A. The IIP consisted of five learning modules, each covering different aspects of the Perfectionism Social Disconnection Model and included an associated workbook. As perfectionism is theorised to lead to psychological distress via interpersonal difficulties and social disconnection (Hewitt et al., 2017), these mechanisms were addressed in the program with the goal of reducing distress that may contribute to overall psychopathology. The program began with a brief introduction and continued with modules on: (1) perfectionism psychoeducation, (2) strategies for managing interpersonal sensitivity, (3) strategies for managing interpersonal hostility, (4) strategies for improving social connections, and (5) a summary of what was learned. These learning modules provided explanations of the specific components in the Perfectionism Social Disconnection Model (Hewitt et al., 2017) and associated management strategies (e.g., mindfulness, self-compassion). The program took 2.5 h to complete over the span of a week and was delivered through PowToon videos on an online e-therapy platform.

The content of the HBMP was developed based on recommended advice from the Australian Government. It consisted of a brief introduction followed by five learning modules covering different aspects of wellbeing and included an associated workbook. These modules covered: (1) healthy body and mind psychoeducation, (2) strategies for managing nutrition, (3) approaches for managing fitness, (4) strategies to improve sleep, and (5) a summary of what was learned. These learning modules were structured to provide information on these aspects of wellbeing, followed by management skills (e.g., meal planning, sleep hygiene). The duration and delivery of the program was identical to the IIP.

Outcomes

Outcome measures were administered at baseline (T0), post-treatment (T1, one month after randomisation), and follow-up (T2, three months after randomisation) via an online survey platform (www.redcap.com). Demographic variables were assessed at baseline only, while acceptability measures were administered at post-treatment only. Participants were reminded to complete assessments up to five times in intervals of 2 days to reduce dropout.

Measures to assess feasibility and acceptability

Feasibility was assessed using retention rates in the trial, operationalised as non-responses to follow-up assessments. To assess acceptability, we gathered data from the e-therapy platform to objectively assess engagement with the program, operationalised as login information and percentage of completed program activities. We also administered the Client Satisfaction Questionnaire adapted to Internet-based interventions (CSQ-I), which measured satisfaction with the online programs using a sum of eight items (Boß et al., 2016). In the current study, the CSQ-I demonstrated excellent internal consistency (total score: 8–32: α = 0.94). Participants also completed a feedback survey, providing information on satisfaction in several domains (e.g., specific modules, the e-therapy platform) and written responses regarding the benefits and challenges associated with engagement with their allocated program. We did not establish a priori thresholds for the feasibility and acceptability criteria.

Measures to assess potential efficacy

We used the Kessler Psychological Distress Scale (K-10; Kessler et al., 2002) as our primary efficacy measure. It was selected as the primary efficacy measure because the intervention aims to primarily reduce the negative consequences of perfectionism, such as psychological distress, rather than perfectionism itself as previous literature has suggested that perfectionistic traits may require longer, more intensive interventions to elicit changes (Smith et al., 2023). K-10 scores range from 10 to 50, with higher scores indicating greater psychological distress. A score between 20 and 24 suggests mild psychological distress, while scores between 25 and 29 indicate moderate distress and scores between 30 and 50 suggest severe distress (Kessler et al., 2003). In the current study, the K-10 demonstrated excellent internal consistencies at each time point (α^T0 = 0.92, α^T1 = 0.92, α^T2 = 0.94).

Our secondary efficacy outcomes included various measures of perfectionism, such as the Hewitt and Flett Multidimensional Perfectionism Scale (HF MPS; Hewitt & Flett, 1991). The HF MPS assessed trait levels of perfectionism, including socially prescribed (α^T0 = 0.84, α^T1 = 0.90, α^T2 = 0.88), self-oriented (α^T0 = 0.84, α^T1 = 0.89, α^T2 = 0.92), and other-oriented (α^T0 = 0.87, α^T1 = 0.87, α^T2 = 0.86) perfectionism, and demonstrated good to excellent internal consistencies (total score for subscales: 15–105). Furthermore, in line with previous intervention research (Egan et al., 2014), the concern over mistakes and personal standards subscales of the Frost Multidimensional Scale (FMPS; Frost & Marten, 1990) were used to assess perfectionistic attitudes. Both the concern over mistakes (total score: 9 − 45; α^T0 = 0.88, α^T1 = 0.95, α^T2 = 0.93) and personal standards subscales (total score: 7 − 35; α^T0 = 0.79, α^T1 = 0.80, α^T2 = 0.89) demonstrated acceptable to excellent internal consistency.

Our other secondary measures included the Rejection Sensitivity Questionnaire (RSQ; Downey & Feldman, 1996) which was used to assess rejection sensitivity (total score: 1 − 36) and demonstrated excellent internal consistencies (α^T0 = 0.92, α^T1 = 0.94, α^T2 = 0.93). The Buss Perry Aggression Questionnaire (BPAQ; Buss & Perry, 1992) assessed interpersonal hostility, demonstrating excellent internal consistencies (total score: 29–145; α^T0 = 0.90, α^T1 = 0.93, α^T2 = 0.94). The UCLA Loneliness Scale (UCLALS; Russell, 1996) was used to assess subjective loneliness (total score: 20 − 80), demonstrating excellent internal consistencies (α^T0 = 0.95, α^T1 = 0.95, α^T2 = 0.95). The Academic Resilience Scale (ARS; Cassidy, 2016) was used to measure resilience to academic adversity (total score: 30–150), demonstrating acceptable to good internal consistencies (α^T0 = 0.74, α^T1 = 0.77, α^T2 = 0.88).

We also included the Self-Compassion Scale-Short Form (SCS SF; Raes et al., 2011) which measured self-compassion (total score: 1–5) and demonstrated acceptable internal consistencies (α^T0 = 0.71, α^T1 = 0.70, α^T2 = 0.75). We used the curiosity and decentring subscales from the Toronto Mindfulness Scale-Trait Version (TMS; Davis et al., 2009) to assess aspects of trait mindfulness. The curiosity (total score: 0–24; α^T0 = 0.90, α^T1 = 0.89, α^T2 = 0.91) and decentring (total score: 0–28; α^T0 = 0.82, α^T1 = 0.85, α^T2 = 0.83) subscales demonstrated good to excellent internal consistency. We also used the Distress Tolerance Scale (DTS; Simons & Gaher, 2005) to assess capacity for tolerating distress (total scores: 15–75), which demonstrated good to excellent internal consistencies (α^T0 = 0.92, α^T1 = 0.89, α^T2 = 0.95). For all secondary outcomes (except the ARS, SCS SF, TMS, and DTS), higher scores indicated greater levels of the respective construct.

Sample size

This study aimed to examine the feasibility and acceptability of a revised version of the previously piloted IIP to provide data for estimating parameters required to design a definitive randomised controlled trial. Thus, a formal sample size calculation was not required. However, Whitehead et al. (2016) provided guidelines for selecting the optimum pilot trial sample size based on effect size to aid in planning a larger, sufficiently powered efficacy trial (90% power and two-sided 5% significance). Based on previous research on web-based interventions for university students, we expect a larger RCT to find small between-group effect size differences on the K-10 (Stallman et al., 2019). Therefore, for an expected small difference between groups (0.1 < d < 0.3), we chose a sample size target of 50 (25 per arm; Whitehead et al., 2016). This was further increased to a target of 70 (35 per arm) to allow for an attrition rate of 20%.

Randomisation

Participants self-enrolled in the study between the 11th to 29th of March 2023, and eligible participants were identified automatically by the survey platform (www.redcap.com). Participants were then randomly allocated to the intervention ‘IIP’ or the attention placebo control condition ‘HBMP’ using a random block allocation per a computer-generated randomised schedule (www.sealedenvelope.com). We used block sizes of 4, 6, and 8. A blinded research assistant generated the random allocation sequence and assigned participants to conditions by providing them with log-in details for their allocated program. No allocation concealment mechanism was employed, as the research assistant had access to the full allocation list. Other members of the research team remained blind to group allocation status until data analysis was completed. Finally, as participants were informed that the study aimed to investigate differences between two wellbeing treatment programs, they were partially blinded to the true research aim of the study.

Statistical analyses

Data analyses were conducted using SPSS v.29. Data were first examined for random responding and assumptions of linearity, homoscedasticity, normality of residuals, and missing data. Participants who answered more than one out of three attention check items incorrectly in a survey (N = 2; n = 1 at post treatment and n = 1 at follow-up) had their data for that time point excluded from the final analysis. Visual inspection of plots of standardised residuals vs. predicted values showed no severe violations of linearity, homoscedasticity, and normality of residuals in the outcome variables. In addition, Little’s Missing Completely at Random test was also conducted and showed no cases that displayed systematically missing data (χ² = 114.757, p = 1.000).

For our primary objective, descriptive and frequency analyses were used to examine our feasibility and acceptability outcomes. Independent samples t-tests were used to compare numerical variables, and chi-square tests were used to compare categorical variables when all cell observations exceeded five. When one or more cell observations were less than five, Fisher’s exact tests were applied for 2 × 2 contingency tables, while Fisher-Freeman-Halton tests were employed for contingency tables larger than 2 × 2. As a measure of effect size, we generated Cohen’s d for numeric tests (0.41, 1.15, and 2.70 represent small, medium, and large sizes, respectively; Ferguson, 2009) and Cramers V or Phi for categorical tests (0.2, 0.5, and 0.8, representing small, medium, and large effect sizes, respectively). Qualitative feedback data for the program was coded in Nvivo 12 by the first author and an independent research assistant using a thematic and deductive approach. Cohen’s kappa percentage agreement was 88.93%, indicating excellent agreement between raters.

For our secondary objective, linear mixed models with random intercepts were conducted to assess the potential efficacy of the program on our outcomes of interest. Linear mixed models were fitted to estimate the fixed effects of group, time, and the group-time interaction, and the random effects of participant (specifying a random intercept to control for individual differences). All analyses included all available observations using maximum likelihood estimation. Satterthwaite approximation degrees of freedom and 95% confidence intervals were reported for comparisons. We also examined clinically significant changes for psychological distress and perfectionism. Cut-offs from previous intervention research were used for psychological distress and the concern over mistakes subscale, which corresponded to criterion B (i.e., scores within the normative range of a non-clinical population; Jacobson & Truax, 1991) of clinically significant change (Rickwood et al., 2015; Rozental et al., 2017). Previous cut-offs were not available for the Hewitt and Flett Multidimensional Perfectionism subscales and the personal standards subscale. Therefore, we calculated these per Criterion C (i.e., scores closer to the non-clinical vs. psychiatric population) and Criterion B, respectively, dependent on values available from past research (Egan et al., 2016; Franco et al., 2014; Hewitt et al., 1991). Chi-square tests, Fisher’s tests, or Fisher-Freeman-Halton tests were used to compare clinically significant change. Last, we conducted exploratory analyses to examine if engagement may have impacted the potential efficacy of the IIP by generating Pearsons r correlations between activities completed in the IIP and change scores in outcome measures (0.2, 0.5, and 0.8 represent small, medium, and large sizes, respectively; Ferguson, 2009).

Participants

At baseline, participants were on average 19.52 years of age (SD = 5.04) and most participants identified as cisgender women (n = 62; 88.60%). Most participants identified their ethnicity as Australian (n = 21; 30.0%), Asian (n = 21; 30.0%), European (n = 12; 17.2%), or Middle Eastern (n = 4; 5.7%), were employed (part-time: n = 20, 28.6% and/or casual: n = 36, 51.4%), and single (n = 47; 67.1%). On average, participants reported experiencing severe psychological distress at baseline (M = 33.23, SD = 8.82). Of these participants, five (7.1%) were likely not psychologically distressed, nine (12.9%) were likely mildly psychologically distressed, seven (10.0%) were likely moderately psychologically distressed, and 49 (70.0%) were likely severely psychologically distressed.

Analyses of feasibility and acceptability

Retention in the trial

Of the 70 randomised, 46 (65.71%) and 34 (48.57%) responded to the post treatment (IIP: n = 21, HBMP: n = 25) and follow-up assessment (IIP: n = 19, HBMP: n = 15), respectively. Only baseline rejection sensitivity significantly differed between responders versus non-responders such that participants who did respond evidenced lower levels of rejection sensitivity (p = .032, d = 0.52; see Table 1). Group allocation did not differ between those who responded to the post-treatment (χ² = 1.014, p = .314, V = 0.12) or follow-up assessment (χ² = 0.915, p = .339, V = 0.11).

Engagement with the program

Of the 70 randomised to the intervention or control group, 60 (85.71%) logged into and accessed their allocated program (IIP: n = 31, HBMP: n = 29). Independent samples t-tests showed that participants who accessed their respective program had higher levels of socially prescribed perfectionism, concerns over mistakes, and rejection sensitivity, and lower levels of self-compassion compared to those who did not access their allocated program (all p’s < 0.05; see Table 2). No other differences were evident. The mean percentage of the activities completed was 36.23% (SD = 31.07) for the IIP and 49.43% (SD = 36.58) for the HBMP, and these differences were not statistically significant (t=-1.627, p = .108, d = 0.39). In addition, comparison tests showed that participants who had engaged with their allocated program (i.e., completed 15% or more of activities) had significantly higher levels of socially prescribed perfectionism and interpersonal hostility, and were more likely to be taking medication at baseline compared to those who did not engage. No other differences were found (see Table 2).

Satisfaction with the program^{Footnote 1}

The average score for the CSQ-I was 24.47 (SD = 4.68) for the IIP and 23.00 (SD = 5.50) for the HBMP, which did not significantly differ (t = 0.948, p = .349, d = 0.29). Our feedback survey for the IIP showed that 66.7% of participants rated Modules 1–3 as most helpful to them, and 77.8% of participants reported that the concepts taught reflected their experience of perfectionism quite a bit or very much so. In addition, 94.5% of participants reported the website navigation to be very or quite easy. For homework tasks, 44.4% reported completing one task, 33.3% completed two tasks, 5.6% completed three tasks, and 16.7% completed all four tasks. However, 50.0% of participants reported that the activities only helped a little bit to somewhat to address their perfectionism, 66.7% of participants reported that the program was a bit or much too long, 38.9% indicated that they experienced barriers to completing the program, and 55.56% of participants reported that the program could benefit from individualised guidance. Last, thematic analyses showed two themes (benefits and challenges) and four sub-themes (internal, external, content, and barriers), which are described in Appendix B.

Analyses of potential efficacy

Intervention effects on efficacy outcomes

Table 3 shows the estimated marginal means, mean changes, and within- and between-group effects of the intervention on our efficacy outcomes. Statistically significant interactions between condition and time were observed for perfectionism scores, interpersonal hostility, loneliness, and self-compassion, favouring the perfectionism intervention group. Our analyses revealed that other-oriented perfectionism showed significant interactions at both post-treatment and three-month follow-up. In addition, changes in socially prescribed perfectionism, concern over mistakes, interpersonal hostility, and self-compassion scores were evident at post-treatment but were not maintained at three-month follow-up. Significant interactions were found for self-oriented perfectionism, personal standards, and loneliness at three-month follow-up but not at post-treatment. No significant interactions were observed between condition and time for the other clinical characteristics. Furthermore, within-group analyses showed significant improvements from baseline for almost all variables except academic resilience and mindful decentring in the IIP group. For these significant effects, all improvements at post-treatment were maintained at three-month follow-up except for psychological distress, hostility, and mindful curiosity. Within-group analyses for the HBMP indicated that only rejection sensitivity and self-compassion showed significant improvements at three-month follow-up from baseline (but not at the post-treatment assessment). All other within-subjects’ effects for the HBMP group were non-significant.

Clinically significant change

Table 4 displays clinically significant change stratified by group allocation. For socially prescribed, self-oriented, and other-oriented perfectionism, there were significant between-group differences on clinically significant change at post treatment (p= .013, p= .037, and p= .038, respectively) but not at three-month follow-up. As shown in Table 4, between-group differences tended to indicate a higher proportion of IIP participants showing positive changes and a smaller proportion showing no change or worsening outcomes compared to HBMP participants. No other between-group differences were evident for psychological distress, concern over mistakes, or personal standards subscales of the FMPS.

Engagement by changes in efficacy outcomes

Table 5 (displayed in Appendix C) displays the associations between the percentage of activities completed and changes in outcome measures. There was a significant small positive correlation between activities completed in the program and changes in other-oriented perfectionism at one-month follow-up (r = .46, p = .035). There were also small correlations between activities completed in the program and changes in perfectionism, rejection sensitivity, interpersonal hostility, loneliness, psychological distress, self-compassion, and distress tolerance at three-month follow-up (r’s = − 0.23-0.41; see Table 5). However, these associations were not statistically significant, which was likely due to a lack of statistical power.

The aim of this trial was to examine the feasibility and acceptability of an online perfectionism program compared to an attention placebo control and to provide data to estimate parameters required to design a definitive randomised controlled trial. Our findings indicated that participants were satisfied overall with the IIP, but also highlighted important issues related to retention and engagement. Moreover, we found that the IIP significantly decreased levels of perfectionism and interpersonal hostility and increased self-compassion relative to the attention placebo control condition, but this was not evidenced for other important clinical characteristics (e.g., psychological distress). However, these results should be interpreted with caution given this study was not sufficiently powered to find statistically significant differences between groups and the retention rate was low. Nevertheless, findings are discussed below with an emphasis on their implications for planning of a future definitive trial to further evaluate the IIP.

A critical area of improvement for future testing of the IIP is the retention rate as the retention rate in our study (65.71% for post treatment and 48.57% for follow-up) was suboptimal. Of note, trials conducted among university student populations often exhibit notably lower retention rates. For example, Ferrari et al.’s (2022) systematic review found retention rates ranging from 29.3 to 88.9% for digital interventions targeting psychological wellbeing in university students. Given that our study found no significant differences in retention rate by group allocation status, retention rate may be at least partially due to population itself. Nevertheless, researchers are responsible for finding user-centered strategies to enhance retention. Systematic reviews have identified several effective methods, such as monetary incentives and reminders to non-responders (text messaging or telephone; Elfeky et al., 2020), which could be utilised in a future definitive trial.

Similarly, a future definitive trial evaluating the IIP should aim to improve participant engagement with the program. On average, participants completed only 36.2% of the IIP, and individuals with lower levels of symptomatology were less likely to access and engage with the programs. This aligns with previous research which has shown that unguided vs. guided internet-delivered interventions tend to have lower rates of module and program completion (Baumeister et al., 2014). In addition, university students may be less motivated to change their perfectionism as they could perceive it to be advantageous in the university context (Stoeber & Hotham, 2013), particularly as the detrimental consequences of perfectionism may have not fully manifested due to their young ages. To improve engagement with online interventions, studies have explored strategies including incorporating explicit messages to boost intrinsic motivation (e.g., sense of control, identification with the program; Borghouts et al., 2021) and using persuasive technology (e.g., providing personalised content; Lipschitz et al., 2023). Moreover, Lipschitz et al. (2023) suggests that heightened severity may motivate engagement which is consistent with our own results, suggesting that the program may be more suited to those with higher symptomology and inclusion/exclusion criteria may need to be modified. Thus, a future definitive trial should adopt strategies and modify the inclusion/exclusion criteria to improve engagement with the IIP.

Moreover, our feasibility study showed some promise regarding the potential effects of the IIP on efficacy outcomes. Our analyses revealed that participants allocated to the IIP demonstrated significantly greater reductions in measures of perfectionism, interpersonal hostility, and loneliness, along with increases in self-compassion, compared to those allocated to the HBMP. These findings are promising in comparison to existing perfectionism interventions that typically involve 8–12 sessions and have not evidenced reductions in socially prescribed and other-oriented perfectionism between treatment and control conditions (for meta-analysis, see Smith et al., 2023). However, reductions in perfectionism may be transitory as clinically significant change was most pronounced at post-treatment but not maintained at follow-up. This might indicate that the brevity of the program and/or low engagement with it was insufficient to produce sustained changes, consistent with previous research suggesting longer and more intensive interventions are required (Smith et al., 2023). Moreover, although we did not observe significant interactions for other clinical characteristics, we found that these still significantly improved over time in the IIP but not in the HBMP group. In addition, increased completion of activities was associated with small to medium improvements in outcome measures, suggesting that our null efficacy results could also be attributed, in part, to low engagement. However, caution is warranted when interpreting the efficacy data from this feasibility trial, as a larger sample size is needed to validate these potential intervention effects.

Furthermore, our analyses highlighted potential modifications to the IIP. For instance, the qualitative analyses showed that some participants commented that some of the modules were not applicable to them. This might be because the PSDM specifies that interpersonal sensitivity and hostility, together, are not necessary for perfectionism to lead to social disconnection. For example, an individual with high levels of socially prescribed perfectionism may only exhibit interpersonal sensitivity but not hostility which could contribute to social disconnection. A potential solution might be that participants can complete a questionnaire to assess if these mechanisms are relevant to them before being directed to complete specific modules, thereby increasing the brevity and relevance of the intervention to its participants.

These findings should be considered in conjunction with the limitations of our feasibility study. First, our retention rate may impact the validity of the results, as high rates of attrition can lead to bias and reduce both power and generalisability. Second, due to our engagement rate, the data on feasibility/acceptability and outcome measures have been provided by participants who only partially completed the program. Third, although our small sample size was appropriate for a feasibility study, our analyses were underpowered to detect reliable changes in our efficacy outcomes. Fourth, almost 90% of participants in this feasibility trial identified as cisgender females, which limits the generalisability of our findings to individuals who identify another way. Fifth, relying on self-report questionnaires to collect outcome data on behavior carries the risk of participant-reported bias. Sixth, we did not collect data on potential harms related to engagement with the IIP, which should be examined in a future definitive trial. As such, these limitations can be overcome by recruiting a larger and more gender-diverse sample, including strategies to improve retention and engagement, as well as by using both subjective and objective (e.g., informant reports, semester results) measures to assess our efficacy outcomes.

In conclusion, the current feasibility study sought to test a brief, accessible, and efficacious online program to assist university students in managing their perfectionism. The study results showed that the IIP was generally well-accepted but improvements to retention and engagement are needed such that participants may have struggled to engage with the intervention, exhibited insufficient application of the skills taught, and may have not internalised the information. Promisingly, we still obtained statistical significance for some of the efficacy outcomes relative to the attention placebo control condition. However, further refinements may be necessary to ensure that other important clinical characteristics are adequately targeted, and that all treatment outcomes evidence sustained changes. Thus, a future definitive trial could benefit from including strategies to improve retention rate, such as offering monetary incentives and using text message reminders. Additionally, modifying the program’s delivery (e.g., personalisation) and expanding the recruitment scope (e.g., a larger and more diverse sample) should be considered to further evaluate the efficacy of the IIP.