What attributes of interventions for osteoarthritis drive preferences? A discrete choice experiment involving cross-sectoral and multi-disciplinary stakeholder groups

Purpose: Evidence-based interventions for managing osteoarthritis (OA) are under-utilised - leading to a missed opportunity for reducing disease burden. Delivery of OA interventions is influenced by choices made by stakeholders such as consumers, providers and policy-makers of OA health care. When stakeholders choose OA interventions, what attributes of the interventions are most important to them? Understanding stakeholders’ preferences in this respect could help unravel the evidence-practice gap and suggest strategies to address it. This study sought to discover the relative importance of the attributes of OA interventions and to evaluate whether stakeholders’ preferences can be explained by their sociodemographic characteristics. Methods: Between November 2017 and July 2018, a convenience sample of OA health care consumers, providers, policy-makers and other OA experts completed an online discrete choice experiment (DCE) based on the PAPRIKA method - an acronym for Potentially All Pairwise RanKings of all possible Alternatives - implemented using 1000minds software. The DCE revealed stakeholders’ weights representing the relative importance to them of eight attributes characterising OA interventions identified in an earlier study of ours: accessibility (Acc), cost of the intervention (Cos), duration of the intervention effect (Dur), effectiveness (Eff), quality of evidence about the intervention (Qua), recommendation for using the intervention (Rec), risk of mild or moderate effects (RMi) and risk of serious harm (RSe). Each attribute comprised 2-4 levels, informed by the Royal Australian College of General Practitioners 2018 OA clinical guidelines, an expert panel consensus or a review of the literature. To assess differences in the mean weight of attributes assigned by different stakeholder groups, the Holm-Šidák corrected Dunn’s pairwise comparison test was used. Fractional multinomial logit (FMNL) was used to evaluate the magnitude of association between the sociodemographic characteristics of the stakeholders against the eight attribute weights. Results: The DCE was completed by 178 people (mean [SD] age 53 [13] years; 114 female): 63 consumers, 79 providers, 24 policy-makers and 12 OA experts. The ranking of the attributes in decreasing order of relative importance (mean weights in parentheses) was: 1st - Rec (0.190), 2nd - Qua (0.176), 3rd - Eff (0.150), 4th - Dur (0.132), 5th - RSe (0.128), 6th - RMi (0.094), 7th - Cos (0.066), 8th - Acc (0.063). Dunn’s test detected significant differences for the Eff, RMi and Qua attributes (Table 1). To account for risk of non-response sample bias, the full-sample mean rank of the attributes was also calculated after adjusting the weight of each group sample to equivalence. After adjustment, the ranking of the attributes was the same except for the Dur and RSe attributes which swapped ranks for 4th and 5th places. Average partial effects of the FMNL model revealed an association between the five attribute weights Dur, Eff, Qua, Rec and RMi, and stakeholder group (Table 2). Specifically, health care providers, on average, place 4.3% more importance on Rec, whereas policy-makers place 4.9% more importance on Qua, and 4.7% less importance on Dur, relative to the other attributes and compared with consumers. OA experts, on average, place 3.4% less importance on RMi, relative to the other attributes and compared with providers, and compared with policy-makers, place 4.5% less importance on Eff, relative to the other attributes. The average partial effects were relatively small (no more than 5.7%) after accounting for other sociodemographic characteristics, corroborating the practically null difference in stakeholder group attribute weights compared to the full sample mean, as similarly observed in the between-group analysis. Conclusions: Stakeholders’ preferences for the attributes of OA interventions are independent of the stakeholder group they belong to and other sociodemographic characteristics. Although some statistically significant associations were detected, the differences were small, as reflected by their overall rankings, and are unlikely to be meaningful in practice. Our findings imply that the preferences of stakeholders responsible for providing, recommending or delivering OA interventions do not meaningfully differ from consumers’ preferences, and align with the evidence-based paradigm. This absence of differences among stakeholder groups implies that preferences are unlikely to be a barrier to implementing evidence-based OA interventions.Table 1Comparison of the stakeholder group mean attribute weights compared to the full-sample meanMean attribute weight (SD), RankGroup (N)Recommendation to use the intervention now (Rec)Quality of the evidence about the intervention (Qua)Effectiveness of the intervention (Eff)Duration of the intervention effect (Dur)Risk of serious harm (RSe)Risk of mild to moderate side-effects (RMi)Cost of the intervention (Cos)Accessibility to the intervention (Acc)Full sample (178)0.190 (0.064)10.0176 (0.064)20.150 (0.062)30.132 (0.73)40.128 (0.070)50.094 (0.060)60.066 (0.041)70.063 (0.055)8Consumers (63)0.185 (0.122)10.156∗∗ (0.097)20.138 (0.107)40.151 (0.156)30.133 (0.128)50.098 (0.116)60.073 (0.084)70.067 (0.110)8Weight difference†-0.005-0.020-0.0120.0190.0050.0040.0070.004Providers (79)0.195 (0.111)10.183 (0.089)20.156 (0.094)30.122 (0.084)50.122 (0.097)40.102* (0.083)60.058 (0.049)80.062 (0.073)7Weight difference†0.0050.0070.006-0.010-0.0060.008-0.008-0.001Policy-makers (24)0.185 (0.158)20.210∗ (0.217)10.172∗ (0.144)30.144 (0.14850.118 (0.142)40.072∗ (0.135)70.076 (0.111)60.053 (0.118)8Weight difference†-0.0050.0340.022-0.018-0.010-0.0220.010-0.010OA experts (12)0.204 (0.298)10.171 (0.249)20.133 (0140)50.134 (0.278)40.165 (0.361)30.069 (0.115)60.058 (0.092)80.066 (0.185)7Weight difference†0.014-0.005-0.0170.0020.037-0.025-0.0080.003Holm-Sidak corrected Dunn's *p<0.05, **p=0.001. †Group minus full sample attribute weight mean Open table in a new tab Table 2Average partial effects (APE) of the fractional multinomial logit modelAverage partial effects†Sociodemographic characteristicsRecommendation to use the intervention now (Rec)Quality of the evidence about the intervention (Qua)Effectiveness of the intervention (Eff)Duration of the intervention effect (Dur)Risk of serious harm (RSe)Risk of mild to moderate side-effects (RMi)Cost of the intervention (Cos)Accessibility to the intervention (Acc)Providers (ref: consumer)0.043∗∗ (0.150)0.016 (0.012)-0.003 (0.015)-0.042∗ (0.019)0.008 (0.018)0.014 (0.014)-0.010 (0.010)-0.009 (0.012)Policy-makers (ref: consumers)0.028 (0.016)0.049∗∗ (0.019)0.018 (0.016)-0.047∗∗ (0.017)-0.019 (0.019)-0.024 (0.018)0.010 (0.011)-0.012 (0.015)OA experts (ref: consumers)0.057 (0.024)0.007 (0.021)-0.029 (0.019)-0.030 (0.026)0.034 (0.026)-0.020 (0.017)-0.009 (0.012)-0.009 (0.017)Policy-makers (ref: providers)-0.014 (0.021)-0.033∗ (0.016)-0.019 (0.013)-0.006 (0.015)-0.011∗ (0.014)0.038∗ (0.017)-0.020∗ (0.008)-0.003 (0.013)OA experts (ref: providers)0.014 (0.021)-0.009 (0.022)-0.026 (0.014)-0.012 (0.022)0.042 (0.022)-0.034∗∗ (0.012)0.006 (0.009)0.000 (0.014)OA experts (ref: policy-makers)0.029 (0.023)-0.042 (0.024)-0.045∗∗ (0.017)-0.018 (0.025)0.052∗ (0.025)-0.005 (0.020)-0.020 (0.011)0.003 (0.018)Female (ref: Male)0.018 (0.012)-0.015 (0.010)-0.009 (0.010)-0.014 (0.013)0.009 (0.011)0.002 (0.009)0.002 (0.006)0.005 (0.018)Australian (ref: New Zealander)-0.278 (0.019)-0.012 (0.015)0.024 (0.015)0.025 (0.015)0.032 (0.019)-0.008 (0.023)-0.012 (0.014)-0.014 (0.015)DHB or MoH employee (ref: other employer)0.026∗ (0.013)-0.003 (0.011)-0.196 (0.013)-0.014 (0.016)-0.001 (0.014)0.018 (0.010)0.004 (0.007)-0.011 (0.009)Age (mean age=54)0.001∗ (0.001)-0.001 (0.000)-0.001 (0.000)-0.000 (0.001)0.000 (0.001)-0.000 (0.001)0.000 (0.000)0.000 (0.000)Work experience (years; mean years experience=16)0.000 (0.001)0.001 (0.001)-0.000 (0.000)-0.000 (0.001)0.000 (0.001)-0.001 (0.000)0.000 (0.000)0.000 (0.000)Standard errors are in parentheses. †Negative coefficents indicate less importance. *p<0.05, **p<0.01. DHB or MoH= District Health Board or Ministry of Health. Separate regressions were run for the Providers and Policy-makers reference categories. p<0.001 ‘goodness-of-fit’ Wald Chi-square for each regression, indicating at least one of the coefficients has a significant impact on the attributes. Open table in a new tab Holm-Sidak corrected Dunn's *p<0.05, **p=0.001. †Group minus full sample attribute weight mean Standard errors are in parentheses. †Negative coefficents indicate less importance. *p<0.05, **p<0.01. DHB or MoH= District Health Board or Ministry of Health. Separate regressions were run for the Providers and Policy-makers reference categories. p<0.001 ‘goodness-of-fit’ Wald Chi-square for each regression, indicating at least one of the coefficients has a significant impact on the attributes.