Background: Lymphoma is usually diagnosed on biopsy, but clues to a possible diagnosis may be found earlier in the diagnostic pathway when patients undergo CT imaging. Often CT scans are reported as showing likely lymphoma, leading to specialty referral before tissue diagnosis. A proportion of these patients have alternative diagnoses on biopsy, with a subsequent delay in referral to the most appropriate specialty. In the UK NHS Delivering Cancer Waiting Times: Good Practice Guide (2014), a top priority from patients for improving their experience was to be told they did not have cancer as soon as possible and, where they did have cancer, for a treatment plan to be made early. With the pressure of cancer waiting times and 2-week wait targets, any factors which can aid in earlier and more appropriate referral for patients will be valuable for health service providers and patients.
Aims: Our aim was to identify factors that may predict a lymphoma diagnosis at the time of CT, prior to diagnostic biopsy. Our long-term goal is to create a diagnostic model which can be used to identify patients who are more likely to have lymphoma and who should be referred early to the appropriate team, leading to a more streamlined diagnostic pathway and improved patient experience.
Methods: Between 1st July 2019-30th June 2021, 441 consecutive patients were identified who had undergone a CT scan reported as showing ‘likely lymphoma’. Patients were excluded if they were <18 years old or had a history of lymphoma. 311 patients had a biopsy with a tissue diagnosis. Training and validation datasets were randomly selected (2:1), balanced for biopsy outcome (lymphoma, non-cancer, other cancer), extent of disease on CT and presence of bulk (defined as mass ≥5 cm). Univariable logistic regression and Receiver Operating Characteristic Area Under the Curve (ROC-AUC) analysis were performed on the training set. Multivariable analysis, model building and validation is being undertaken.
Results: Our patient cohort were 45.1% female with median age 67 years. Of the 311 patients with a tissue diagnosis, 48.9% (152/311) had a final diagnosis of lymphoma; 24.1% (75/311) had a diagnosis of another malignancy. Non-malignant diagnoses were heterogeneous and included granulomatous disease including sarcoid (19/311) and infection/reactive causes (30/311). There were 208 and 103 patients in the training and validation sets; demographics were well-balanced.
Univariable analysis was undertaken (Table 1) with outcome lymphoma/non-lymphoma. Factors significantly associated with lymphoma diagnosis were all radiological: size of largest lesion (odds ratio, OR, for a 1cm increase: 1.21 (1.09–1.35), p<0.001), presence of bulk (OR 3.29 (1.63–6.64), p=0.001) and para-aortic lymphadenopathy (OR 2.92 (1.57–5.42), p = 0.001). No blood parameters were significantly associated; raised LDH showed some predictive ability for cancer vs not but could not discriminate between other cancer and lymphoma.
Image:
Summary/Conclusion: In univariable analysis, we found parameters associated with an increased chance of a biopsy-proven lymphoma diagnosis when lymphoma was reported as likely on a CT scan. These factors alone showed only moderate predictive ability. Our next step is to assess their role in a multivariable model and test their predictive efficacy within our validation dataset. We plan to present this data at the EHA meeting. If successful, we hope to use this diagnostic model to streamline the diagnostic pathway for patients presenting with a scan suggestive of lymphoma, improving the efficiency of the patient pathway and improving patient experience.