3.4 | Follow-up Care Analysis
Of the 865 survivors included in the final analytic cohort, 282 (32.6%) were not seen in any oncology-related subspecialty clinic (PHO, PNO, PBMT, Medical Oncology) five to seven years after the initial diagnosis. Factors associated with follow-up included risk strata (0.001), age (p=0.008), primary diagnosis (p<0.001), and race/ethnicity (p=0.010). Risk strata was the primary association of interest for this analysis. Low risk patients yielded the highest follow-up with 78% of patients, followed by high risk patients at 70%, and intermediate risk patients at 63%. To include spatial data elements, we restricted analysis to survivors within NC, SC, and VA (n=787). Similar associations with follow-up patterns were observed as well as lower ADI national percentile (p=0.011) and distance to primary treatment site (p<0.0001), though not distance to a COG-affiliated site (p=0.729) (Table 2). Manual chart review of a 10% randomized sample of patients lost to follow-up from each risk strata (n=37) revealed that most patients were instructed to follow-up; however there was no additional follow-up documented in EHR. One low risk, four intermediate risk, and no high risk patients had documentation of care transferred to another institution.
We first built a logistic model to test the unadjusted association between risk stratification and appropriate follow-up care through a likelihood ratio test (LRT). There was strong evidence (p<0.001) that there was an association between risk stratification and follow-up care (Table 3). Pairwise comparisons showed that the odds of receiving follow-up in the five to seven year window after initial diagnosis in the intermediate risk strata was half the odds of receiving follow-up in the low risk strata (OR 0.482; CI 0.3, 0.774, p<0.001). However, there is insufficient evidence to suggest the odds of follow-up is different in pairwise comparisons between high and intermediate risk and high and low risk survivors (p=0.41 and p=0.23, respectively). We then built a multiple logistic regression model adjusted for diagnosis of ALL, gender, age at diagnosis, race/ethnicity as potential confounders. This attenuated the observed association and, after controlling for potential confounding, there was insufficient evidence to suggest there is an association between risk strata and follow-up care (p=0.17) (Table 4).
To test the hypothesis that risk strata may have a different effect if the patient is closer to the primary treatment center (i.e. Duke), we created a model that included the interaction between local residence and risk strata and used a lack of fit test to look for evidence that living in NC, SC or VA modified the effect of risk strata. There was insufficient evidence (p=0.14) to suggest that “local” patients modified the risk strata effect on likelihood to follow-up in the five to seven year window. For the survivorship cohort limited to NC, SC, and VA, we then constructed a multiple logistic regression model to adjust for ALL, gender, age at diagnosis, race/ethnicity as well as ADI, distance to primary treatment center, distance to COG-affiliated site, and RUCA. This yielded similar results with attenuation of the initially observed associated between follow-up care and risk strata, thus there was insufficient evidence to suggest an association between risk strata and follow-up care in the five to seven year window after the initial diagnosis (p=0.11) (Table 4).