Excess mortality by patient’s multimorbidity, socioeconomic, and healthcare factors, among Diffuse Large B–cell and Follicular lymphomas patients in England: A population-based multilevel study


 Background

Socioeconomic inequalities of survival in patients with non–Hodgkin lymphoma (NHL) persists, which may be explained by patients’ comorbidities. We aimed to assess the association between co/multimorbidity and survival in patients diagnosed with Diffuse Large B-cell (DLBCL) or Follicular lymphoma (FL) in England accounting for other socio-demographic characteristics.
Methods

Population-based cancer registry data was linked to Hospital Episode Statistics. We used a flexible multilevel excess hazard model to estimate 5–year net survival and excess mortality by patient’s multimorbidity and comorbidity status adjusted for sociodemographic, economic, healthcare factors, and accounting for the patient’s area of residence.
Results

Overall, 15,516 and 29,898 patients were diagnosed with FL and DLBCL in England between 2005–2013, respectively. Respectively, those with comorbidities and multimorbidities had 1.3 (95% Confidence Interval -CI-: 1.20–1.40) and 1.5 (95% CI 1.27–1.87) times higher excess mortality compared to those without comorbidities. Patients in most deprived areas showed 26% (95% CI 1.20–1.32) excess mortality risk compared to those in least deprived areas.
Conclusion

Co/multimorbidities are consistently associated with poorer survival among patients diagnosed with DLBCL or FL. Comorbidities and multimorbidity need to be considered when planning public health interventions targeting haematological malignancies in England.


Introduction
Non-Hodgkin lymphoma (NHL) is a heterogeneous group of malignancies, and is currently the 6 th most commonly diagnosed cancer in England: in 2014, approximately 32 males and 23 females per 100,000 people were diagnosed. 1 Agestandardised mortality rates of males and females are 12 and 8 per 100,000 person-years, respectively. 1 The heterogeneity in morphology leads to variation in survival probability; for instance, 5-year survival of Follicular Lymphoma (FL) (86.3%) is higher than Diffuse Large B-cell Lymphoma (DLBCL) (54.8%). 2 The healthcare system in England aims to offer equitable access to care for all patients. However, variability in health outcomes amongst patients with similar cancers and sociodemographic characteristics still occur; 2-4 convincing reasons for the variability remains a topic of interest. In 2001, the National Health Service (NHS) Cancer Plan 5 recognised, and aimed to reduce, the disparities in survival. Since implementation, there is no evidence the Plan had an impact on the inequalities. 6,7 The deprivation-gap in survival is still apparent, despite the Plan and successive policies, 5,[8][9][10] illustrating the little understanding of the mechanisms underlying these inequalities and raising the concern that these policies have missed the relevant targets.
Patients' comorbidity status may impact timely diagnosis, possibly leading to treatment with more adverse effects; 11 comorbidities are, on average, more prevalent and severe amongst more deprived patients. 12 However, recent evidence indicates that comorbidity explains little of the differential cancer survival between socioeconomic groups. [13][14][15] Additionally, research suggests that variations in healthcare access, such as location of residence, could partly explain the inequalities. [16][17][18][19][20] Overall, the association between comorbidity and survival in patients with DLBCL and FL, accounting for other sociodemographic characteristics and the area of residence, remains unclear. We hypothesize that the presence of comorbidities is associated with poorer survival independently of patients age and socioeconomic status. We aim to study the association between comorbidities and net survival amongst DLBCL or FL patients accounting for sociodemographic and economic factors in England.

Study design, participants, data and setting
The outcome of the study was the time to death or censoring among DLBCL and FL patients 5 years after cancer diagnosis. We were interested in the estimation of net survival to then derive the excess mortality only due to cancer. Hence, we used England life tables strati ed by deprivation, sex, age and calendar year (2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013) to account for the overall mortality rate from the background population. 26 Comorbidity status was the main exposure. We de ned comorbidity as the existence of other chronic medical disorders, in addition to cancer, the primary disease of interest, which are causally unrelated to the primary disease. 27,28 Records from HES were used to identify patients' comorbidity status based on a computational algorithm published elsewhere. 29 The algorithm seeks for the presence of comorbidities retrospectively and de nes a time window of 6 to 24 months prior to cancer diagnosis where comorbidities are not recorded to avoid bias due to the presence of comorbidities related to cancer (i.e., cardiological comorbidities due to DLBCL or FL cancer treatment). Patient's comorbidity status was adapted from the original Charlson comorbidity index 30 (CCI). We used the Royal College of Surgeons (RCS) modi ed Charlson Score (Supplementary Table S1). 31 The score removes patients with a previous malignancy to avoid bias, does not weight differently comorbidities, and categorises comorbidities as: no comorbidities, one comorbidity and two or more comorbidities (multimorbidity).
Socio-demographic and economic characteristics were collected from the HES dataset. Ethnicity was recorded as white or other. Area-level deprivation, classi ed into one of ve quintiles, was determined by the Index of Multiple Deprivation 32 (IMD), which was based on the Lower Super Output Area 33 (LSOA) residence of the patient at the time of cancer diagnosis. LSOA is a geographical location with a median of 1500 inhabitants. We also include the information regarding patients' diagnosis path (route to diagnosis), a UK speci c programme, classi ed as: accident and emergency room diagnosis, general practitioner referral (routine and urgent referrals where the patient was not referred under two-week-wait), two-weekwait (urgent GP referral with a suspicion of cancer), and secondary care diagnosis (other outpatient and inpatient elective routes). 34

Statistical Analysis
We tabulated the sociodemographic characteristics by DLBCL and FL. To derive patients excess mortality we used a multilevel excess hazard model (EHM) consisting of a smooth function of the baseline hazard ( ) to derive the net survival estimate, and accounting for heterogeneity across LSOAs via the inclusion of a random intercept with mean zero. 35 The statistical contribution of the random effect to the overall goodness of t of the model was tested using a likelihood ratio test statistic with a Chi-square mixture distribution. 36 The was modelled by a cubic B-spline function of time with two knots placed at 1 year and at 3 years after diagnosis where the hazard plateaus.
We include in the model the following variables: age, sex, comorbidities, deprivation, lymphoma subtype, ethnicity, route of cancer diagnosis, and the overall mortality rate from the background population by levels of deprivation, sex, age and calendar year, included as an offset in the model. We included age as a smooth function consisting of a standardized cubic B-spline centred on 70 years and with one knot at the same age. Furthermore, we assumed a time-dependent effect of age at diagnosis, de ned as a B-spline function of time, and represented by the interaction between time and age. The parameter estimates for the variables were interpreted conditionally on the random effect, i.e., they have a cluster-speci c interpretation, where a cluster refers to a given LSOA. From the model we derived the excess mortality hazard ratios (EMHR) and their respective 95% con dence intervals (CI) for all the categorical variables, and the variance of the random effect for the LSOA.

Missing data analysis
We explored the missing data mechanism for each of the three variables with missing data. Due to clustered data and partially observed categorical variables, we used the latent normal joint modelling multiple imputation approach, under a missing at random assumption (MAR). The imputation model included all fully-and partially-observed variables, vital status indicator, the Nelson-Aalen estimate of the cumulative hazard, and accounted for clustering of patients within lower-super output areas. We generated 10 imputed datasets. The multilevel EHM was tted to each of these datasets, and results combined using Rubin's rules. 37,38 Overall tests for the effects of age after multiple imputation were done using the F-based procedure for the test of multiple parameters after multiple imputation. 39 We used R software for all data analyses; the mexhaz 35  comorbidity score of 1, and 2,142 (4.7%) had a comorbidity score of 2 or more. The prevalence of at least one comorbidity was higher amongst DLBCL (10.7%) compared to FL (7.5%). The mean age was 66.2 years overall. The average age was lower amongst FL compared to DLBCL, 63.9 compared to 67.4 years, respectively. The prevalence of DLBCL was higher amongst deprived areas (16.0%) than FL (14.4%). 'White' was the most prevalent ethnicity (94.3%) compared to minority (5.7%). GP referral was the most common route to diagnosis amongst FL (44.0%); whereas, amongst DLBCL, A&E was most common (33.8%). Variables with missing data were ethnicity (29.6%) and route to diagnosis (6.8%). In the multivariable analysis (  (Table 2). Using a mixture chi-squared test, there was strong evidence (p<0.001) that including the random effect improved the t of the model.  Figure 1 shows the increasing EMHR by age at 3 months, 1-and 5-years ( Figure 1A). The EMHR was higher at 3 months compared to 1 or 5 years since diagnosis, particularly among older patients ( Figure 1A). However, within the rst 6 months after diagnosis, the EMHR of older and younger patients were markedly different compared to those aged 70 years ( Figure  1B), which then stabilises, but the increases from 2 years onwards.
Patients without comorbidities had higher net survival compared to those with comorbidity or multimorbidity ( Figure 2). Furthermore, patients living in the most deprived areas had a lower net survival at 5 years since diagnosis compared to least deprived areas. The survival gap for deprivation (deprivation-gap) in net survival was apparent from approximately 3 months after diagnosis regardless the comorbidity status.

Discussion
Comorbidities and multimorbidities are independent risk factors associated with an increased excess mortality risk among DLBCL and FL patients in England. We found strong evidence of a higher excess mortality amongst DLBCL and FL patients diagnosed with comorbidity and multimorbidity compared to patients without comorbidities after adjusting for age, deprivation level, ethnicity, route to diagnosis and accounting for the patient's area of residence. We also found a noticeable deprivation gap in survival, that was consistent regardless of the patient's comorbidity status.
Differences in access to treatments, or risk of adverse effects, may explain some of the disparities in survival among DLBCL and FL patients. Immunotherapies (rituximab) for the treatment of aggressive lymphomas (e.g. DLBCL) was shown to be effective for those of an advanced age. [41][42][43] Rituximab is often used in combination with doxorubicin, an increase in dosage of which is associated with an increased incidence of adverse effects (cardiotoxicity), such as congestive heart failure. 44 Guidelines based on National Institute for Health and Care Excellence (NICE) and the European Society for Medical Oncology (ESMO) both recommend that patients at risk of cardiotoxicity, or low tolerance of intensive therapy, consider a less-intensive treatment regimen. [45][46][47] This less-intensive treatment allocation may partly explain the inequalities in survival by comorbidity status we found.
Evidence shows that the excess mortality among older patients compared to younger patients is highest within the rst 3 months and just after 4 years since diagnosis, which could be because of histological transformation from indolence to aggressiveness. Studies suggest the risk of histological transformation increases by 3% per year. 48 Therefore, the excess hazard amongst older patients may have increased rather than decreased because histological transformation complexi es the treatment and management of the disease.
The importance of understanding the association of comorbid conditions on cancer patients' outcomes has been well documented. 49 To our knowledge, this is the rst study of England cancer registry data that investigates survival by comorbidity status among DLBCL and FL patients. Our results are consistent with previous ndings from a Danish study that showed the hazard of death increased with severity of comorbidity status; 50 however, the Danish study did not account for missing data, and the association with comorbidities was potentially overestimated.
Authors have suggested ways to improve outcomes for patients diagnosed with comorbidities, which include: novel treatment strategies, 51 inclusion of elderly patients in clinical trials, 52,53 and investigation of dose-allocation amongst those with higher comorbidity scores. 54 The deprivation-gap in survival persists even after accounting for prognostic factors such as comorbidity. 50,55,56 Smith et al. 3 reported no variation of the deprivation-gap in survival; however, this study may have lacked power, and used an estimator of net survival that is less consistent than other approaches that were readily available at the time of publication. 57 Consistent with previous studies, 4 survival after GP referral (non-emergency) diagnosis is signi cantly better compared to A&E. However, our study also nds that patients diagnosed through TWW, who would be expected to have worse symptoms and survival, showed no signi cant difference in survival compared to GP referral. There are two possible reasons for the absence of a difference in the effects. Firstly, GPs could advocate for a prompt referral even though the patient is not on the TWW pathway: resulting in patients with similar access to healthcare facilities. Secondly, on the other hand, patients referred through the TWW pathway have more severe symptoms and expected to have a higher excess hazard. Our results show no difference in the excess mortality indicating that TWW pathway prevents patients with more severe symptoms from having a higher excess hazard: suggesting the performance of TWW pathway is at least as bene cial to a patient's survival as GP referral. However, further factors associated with the interactions between comorbidities and health care systems leading to poorer survival among DLBCL and FL cancer patients need to be studied.
The strengths of this study are that, rstly, we used a large population-based sample size obtained from cancer registry databases linked to HES, which encompasses all patients in England with a diagnosis of DLBCL and FL between 2005 and 2013. Furthermore, we used a latent normal joint modelling multiple imputation to treat missing data in ethnicity and route of diagnosis. . This approach allows imputation of a mix of variable types, while accounting for multilevel structures arising from clustering of patients. 39,58,59 Survival at 1-and 5-years since diagnosis of DLBCL and FL in England trails that of other European countries; 60 however, restricting estimates to those surviving at least 1 year after diagnosis (conditional survival) shows a comparable 5-year survival. 61 This indicates that long-term survival differences are largely explained by the increased short-term mortality.
Understanding long-term survival in DLBCL and FL is more complex due to the histological transformation of indolent lymphomas, which would require an adaptation of the treatment, support and management from healthcare facilities. This adaptation could be compounded by the patient's susceptibility to cardiotoxic treatments. Further studies could focus on the mechanisms and inequalities of short-term mortality, long-term survival of patients with transformed lymphomas, and survival of patients at risk of cardiotoxicity.

Conclusion
After accounting for sociodemographic factors, healthcare factors, socioeconomic deprivation and the patient's area of residence, comorbidities and multimorbidity were consistently associated with poorer survival and an increased excess mortality among patients diagnosed with Diffuse Large B-cell (DLBCL) or Follicular lymphoma (FL) in England. Furthermore, survival inequalities in patients with DLBCL and FL persists between socioeconomic levels after accounting for the presence of comorbidities and multimorbidities. These results show the need for the current framework of the National Health Service to improve the survival of DLBCL and FL patients in the most deprived areas of England, but additionally patient management systems need to consider speci c programs for the management of DLBCL and FL patients with comorbidity and multimorbidity.