Likelihood of sharing health data
Across the whole sample there was a high level of willingness to share both mental (89.68%) and physical (92.75%) health data when measured on a binary (yes/no) scale. A McNemar’s test demonstrated that this difference was significant (χ2(1) = 47.67, p < .001). A similar pattern was seen when dividing the sample by individuals who had experienced a mental health condition and individuals who had never experienced a mental health condition (Fig. 1); both participants with experience of mental illness (χ2(1) = 20.02, p < .001), and participants without experience of mental illness (exact McNemar’s test p = .001) were significantly more likely to share physical health data than mental health data.
We next examined the reported likelihood (measured on a Likert scale) of sharing mental/physical health data (Fig. 2). A Wilcoxon signed-ranks test indicated that participants were significantly more likely to share their physical health data than their mental health data, z = − 8.621, p < .001. This effect was also present amongst those with experience of mental illness, z = − 6.412, p < .001.
Two Mann-Whitney U tests were run to determine if likelihood of sharing mental or physical health data differed between the group of people with experience of only mental illness (n = 468) and the group of people with experience of only long-term physical illness (n = 244). Distributions of the responses for the two groups were similar, as assessed by visual inspection. There was no statistically significantly difference between people with only mental illness and people with only physical illness, for either likelihood of sharing mental health data U = 55,776.50, z = − 0.105, p = .917 or likelihood of sharing physical health data U = 55,474.00, z = − 0.327, p = .744.
Factors influencing likelihood of sharing health data
Influence of demographic factors (model 1)
The full ordinal logistic regression model examining the relationship between demographic factors and likelihood of sharing mental health data (Model 1a, Table 2) significantly predicted the outcome over and above the intercept-only model χ2(24) = 68.80, p < .001. The assumption of proportional odds was met χ2(72) = 74.97, p = .38. All variance inflation factor (VIF) values were well below 10, with the largest being 2.75 (for the “Paid” variable), indicating that there was no multi-collinearity. There was no significant difference between people who had and had not experienced a mental illness in the likelihood of sharing mental health data. Those who said they would “prefer not to say” if they had ever experienced a mental illness were much less willing to share their mental health data than people who said they had never had a mental illness (OR 0.23, 95% CI 0.13 to 0.43). An increase in self-rated physical health was associated with an increase in the odds of sharing mental health data (OR 1.15, 95% CI 1.03 to 1.29). An increase in self-rated mental health was marginally associated with an increase in the odds of sharing mental health data (OR 1.11, 95% CI 1.00 to 1.24).
The full ordinal logistic regression model examining the relationship between demographic factors and likelihood of sharing physical health data (Model 1b, Table 2) significantly predicted the outcome over and above the intercept-only model χ2(24) = 87.17, p < .001. The assumption of proportional odds was not met χ2(72) = 105.24, p = .006. As such, an additional adapted analysis which met the assumption was run, giving a similar outcome (Supplementary material 2). Although there was no overall relationship between highest completed level of education and likelihood of sharing physical health data, Wald χ2(5) = 10.22, p = .069, the contrast between the subcategories of postgraduate degree and vocational qualification was significant, such that people with vocational or college-level qualifications were less likely to share their physical health data than people with a postgraduate degree (OR 0.62, 95% CI 0.46 to 0.84). The odds of people who said they would “prefer not to say” if they had ever experienced a mental illness being willing to share their physical health data were four times lower than that of people who said they had never had a mental illness, (OR 0.25, 95% CI 0.14 to 0.47). There was no significant difference between people who had and had not experienced a mental illness in likelihood of sharing physical health data. An increase in self-rated physical health was associated with an increase in the odds of sharing physical health data (OR 1.13, 95% CI 1.01 to 1.27). An increase in self-rated mental health was also associated with an increase in the odds of sharing physical health data (OR 1.18, 95% CI 1.05 to 1.31).
Influence of frequency of NHS use (model 2)
Two cumulative odds ordinal logistic regression analyses with proportional odds were run to determine the association between frequency of mental/physical NHS health care and willingness to share (a) mental and (b) physical health data (Models 2a and 2b, Table 2). The full model (Model 2a, Table 2) significantly predicted the likelihood of sharing mental health data over and above the intercept-only model χ2(4) = 10.55, p = .03. The assumption of proportional odds was met χ2(12) = 19.98, p = .07. All variance inflation factor (VIF) values were well below 10, with the largest being 1.33, indicating that there was no multi-collinearity. An increase in self-rated physical health was associated with an increase in the odds of sharing mental health data (OR 1.18, 95% CI 1.04 to 1.34). Likelihood of sharing mental health data was not predicted by frequency of NHS mental or physical health care, or self-rated mental health.
The full model (Model 2b, Table 2) also significantly predicted likelihood of sharing physical health data over and above the intercept-only model χ2(4) = 22.48, p < .001. The assumption of proportional odds was met χ2(12) = 12.00, p = .45. People who received NHS physical health care more than monthly were more likely to share their physical health data than people who received less frequent care (OR 1.42, 95% CI 1.05 to 1.91). An increase in self-rated physical health was associated with an increase in the odds of sharing physical health data (OR 1.27, 95% CI 1.12 to 1.45). There was no association between frequency of mental health care or self-rated mental health and the likelihood of sharing physical health data.
Influence of satisfaction with the NHS (models 3 and 4)
Spearman’s rank-order correlations were run to assess the relationship between satisfaction with the NHS and likelihood of sharing mental and physical health data. Higher levels of satisfaction with the NHS were associated with greater willingness to share mental health data, rs (1544) = .13, p < .001, 95% CI 0.07 to 0.18, and physical health data rs (1544) = .13, p < .001, 95% CI 0.08 to 0.18. Mann-Whitney U tests were used to examine whether participants with and without experience of mental illness differed in their satisfaction with the NHS. It was found that participants with experience of mental illness were less satisfied with their overall (mean) experience with the NHS (U = 147,117.50, z = − 12.47, p < .001), as well as less satisfied with their first contact for a physical health condition (U = 46,305.00, − 5.52, p < .001) and less satisfied with the physical health care they had received in the previous 12 months (U = 160,531.00, z = − 5.88, p < .001).
To disentangle the relationships between these aforementioned variables, a cumulative odds ordinal logistic regression with proportional odds (Model 3, Table 2) was run to determine the relative associations between the independent variables mean NHS satisfaction and experience of mental illness, and the dependent variable of likelihood of sharing mental health data (with self-rated mental and physical health included as covariate factors). The full model significantly predicted the likelihood of sharing mental health data over and above the intercept-only model (χ2(5) = 66.44, p < .001). The assumption of proportional odds was met (χ2(15) = 21.45, p = .12). All variance inflation factor (VIF) values were well below 10, with the largest being 1.45, indicating that there was no multi-collinearity. In keeping with previous analyses, satisfaction with the NHS (Wald χ2(1) = 20.64, p < .001), and experience of mental illness were associated with likelihood of sharing mental health data (Wald χ2(2) = 29.92, p < .001). Participants who would “prefer not to say” whether they had a mental illness were much less willing than people who said they had never had a mental illness to share their mental health data (OR 0.27, 95% CI 0.15 to 0.49). This means that although experience of mental illness was not related to willingness to share mental health data when examining demographic factors only (Model 1a), accounting for the variable of NHS satisfaction (by including it as a factor in Model 3) revealed a significant relationship between experience of mental illness and higher likelihood of sharing mental health data (OR 1.30, 95% CI 1.02 to 1.62).
Following this, a cumulative odds ordinal logistic regression with proportional odds was run to determine the association between two specific measures of NHS satisfaction (satisfaction with first contact for mental health care, and satisfaction with mental health care in the previous 12 months) on likelihood of sharing mental health data amongst people who have had a mental illness (Model 4, Table 2; n = 480). Ratings of mental and physical health were included as co-variates, as they significantly predicted willingness to share mental health data in a previous analyses (Model 1a). The assumption of proportional odds was met, (χ2(12) = 7.604, p = .82). All variance inflation factor (VIF) values were well below 10, with the largest being 1.33, indicating that there was no multi-collinearity. The final model did not significantly predict likelihood of sharing mental health data over and above the intercept-only model, χ2(4) = 8.885, p = .064. However, an increase in satisfaction with first contact for mental health care was associated with an increase in the odds of sharing mental health data (OR 1.16, 95% CI 1.00 to 1.34). There was no association between satisfaction with previous 12 months’ mental health care, rating of physical health or rating of mental health and willingness to share mental health data.
Influence of specific mental health conditions
A cumulative odds ordinal logistic regression with proportional odds was run to determine the relationship between each primary reported mental health condition and likelihood of sharing mental health data (Table 3). The full model significantly predicted the outcome over and above the intercept-only model (χ2(15) = 43.06, p < .001). The assumption of proportional odds was met (χ2(45) = 52.76, p = .20). In comparison to participants with no experience of mental illness, participants whose primary experience of mental illness was with depression, OCD, personality disorder or bipolar disorder were more willing to share their mental health data (Table 3). There were no cases where experience of a mental illness significantly reduced willingness to share mental health data.