DISCUSSION
Although there were several breakthrough innovations in treatment, such as target therapy, or immune checkpoint inhibitors, oral cancer patients with regional recurrence incurred grave outcomes.2,4Besides radical resection of primary tumor with adequate margin, the next step was to perform an adequate neck dissection. Previous studies had proved the positive association between LN yield and survival rates.8,16 However, strategies to improve LN yield was not clearly found in literatures. This study was the first to explore the impact of LNPA among surgeons for quality of neck dissection. Through continued education, promotion of importance of LN yield in neck dissection surgery, identification of landmarks during surgery and public announce of LN yield average to surgeons might promote the quality of neck dissection and reduce regional recurrence in turn through a difference-in-differences analysis. This soft and non-mandatory method could be applied to all head and neck surgeons widespread.
This study had several strengths. Using a difference-in-differences analysis, the impact of policy or intervention change that was implemented to some groups could be evaluated.13 In this study, oral cancer patients were treated by two departments in our institute, and LNPA was launched among target group in period after 2015 Oct. This situation provided us an opportunity to evaluate the possible positive effect of LNPA on LN yield and regional recurrence through a difference-in-difference analysis. Our finding of this study provided evidences to support the positive effect of LNPA. In difference-in-difference analysis, the coefficient for target group in period after 2015 Oct was 13.4 (p<0.001). In regional recurrence analysis, patients with cN0 disease might get borderline benefit from LNPA with a reduction of regional recurrence rate of 9.4% (p=0.087).
Although the study design of difference-in-differences analysis provided us a chance to explore the effect of policy or intervention at different times, several assumptions should be met. The most common assumptions were the common trends assumption and strict exogeneity. The common trend assumption meant that the unmeasured variables were time-invariant group character or time-varying factors with group invariant attributes.13 It could be examined by checking the time-series graph with a set of parallel lines. Supplementary Figure illustrated the changes of LN yield in different groups and periods. The lines looked like parallel lines. Using matching techniques that could eliminate possible selection bias between different groups and time periods might help the difference-in-difference study, like the LNPA, to reach the requirement of strict exogeneity.15 Other matching method, such as refined covariate balance could also be used.17
All national and international guidelines reported adequate LN yield could significantly improve survival in oral cancer patients, but there is no recommend intervention in clinical practice. Recently, interventions based on the theories of behavior economics are summarized by the acronym NUDGE and has been explored in the healthcare field.18-20 Penn Medicine, for example, has used this strategy widely. Changing the default in the choice of medicine greatly increased the rate of prescriptions for generic medicine.19 Using an active choice alert system in the electric medical record increased the influenza vaccination rate 37% in adults suitable for vaccination.21 Ayala et al. also reported that moving from an ”opt-in” to an ”opt-out” system greatly increased the rate of providing aspirin prophylaxis for preeclampsia prevention.22 Automated dashboard with active choice and peer comparison performance feedback to physicians had increased statin prescribing for primary care physicians.23 In our institute, several feasible and low-cost strategies were applied. At first, weekly conferences were used to explain the association between LN yield and recurrence since 2015 Oct. Thereafter, we encouraged the surgeons to confirm important landmarks during neck dissections came. Furthermore, the average number of LN yield was announced per month in order to give feedback to the surgeons. Multiple NUDGE-like interventions helped the target group to improve the cervical LN yield in oral cancer surgery.
There are several limitations in this study. Although we introduced several interventions to improve the quality of neck dissection in our institution, the study was not completely prospective designed. Second, we didn’t perform the survival analysis due to short-term of follow-up period in those treated in period after 2015 Oct. The minimal follow-up period in our series was 18 months, which represented the least follow-up period for monitoring recurrence, and most of the recurrences could be captured. Furthermore, we used the concept of regional recurrence density, which meant the proportion of regional recurrence among those with neck dissection during a specific period. Figure 3 illustrated the change of regional recurrence density among the target group and control group. Through case matching with propensity score by clinical tumor and node category, the comparison between the two groups were more reasonable. Third, the spillover effects of LNPA to the control group was not estimated. Although it was possible, the effect might be minimal because these two departments located in different floors in our institute.