Variation in guideline adherence in non-Hodgkin’s lymphoma care: impact of patient and hospital characteristics

This study demonstrated substantial hospital variation in guideline adherence for NHL care. Fifteen characteristics at the patient level could partly explain this variation, such as extranodal involvement, multidisciplinary consultation, tumor type, therapy used and hospital region. Hospital characteristics contributed less to the variation in adherence than patient, tumor and professional related characteristics. Patients’ age was involved most frequently as determinant, illustrating that older people are less likely to receive NHL care as described in the guidelines.

Our study showed large gaps between daily practice performance and care as described in the evidence-based guidelines. Large variation in guideline adherence between hospitals is often associated with lower quality of care, since guidelines aim to assist professionals to deliver the most optimal care. However, less adherence does not always indicate lower quality of care: complying with patient preferences or performing less diagnostics due to a low performance status can also point towards patient-centered, safe and deliberately delivered care. It is believed that variation due to deliberately deviate from guidelines is reflected in the upper 10 % of QI scores (90–100 %). Therefore, many studies indicate room for improvement if guideline adherence, as measured by indicators, is below 90 % [9, 22–24]. In our study, 18 out of 20 QIs showed room for improvement, of which 12 QIs demonstrated high hospital variation (>50 %), indicating other factors than patient preferences or performance status might play a role. Similar to our study, Weeks et al. [27] found high variation in NHL management decisions, for example in performing a PET-scan (range 38–95 %) or a bone marrow biopsy (range 21–99 %). Studies concerning other tumor types also showed variation in delivered care between hospitals [18, 28–31].

While this is the first study to investigate determinants at patient as well as hospital level for guideline adherence, and indirectly for the quality of care for NHL patients, other studies examining multilevel determinants have been carried out in several areas, including lung, prostate and (colo)rectal cancer [18, 29, 32–34]. Schroeck et al. [33] provided insight into adherence to QIs for prostate cancer and its regional variation. Most measures showed low adherence rates and high regional variation, for example 72 % variation in follow-up with radiation oncologists (range 14–86 %). They showed that characteristics such as age, clinical stage and number of urologists explained the differences for 5–20 %. Etzioni et al. [32] showed that characteristics as higher-volume surgeons and teaching hospitals contributed to long-term survival in rectal cancer patients, whereas Sacerdote et al. [34] found several social, clinical and hospital characteristics to be associated with the treatment of colorectal cancer, for example, age, gender, hospital volume and an in-hospital radiotherapy service. Mathoulin et al. [29] investigated the quality of colorectal cancer surgery and found several associations with patient, tumor and hospital related factors, such as age, disease stage and hospital type. Finally, Ouwens et al. [18] found patient characteristics to have a greater influence on quality of integrated care than professional or hospital characteristics for patients with non-small cell lung cancer.

Several determinants of guideline adherence and NHL care were found in our study as well. Regarding patient factors, especially patients’ age appeared to influence variation in guideline adherence for NHL care most. For older patients, it can be argued that suboptimal diagnostics and suboptimal but better tolerated therapies sometimes are the best achievable care. However, the reasons for deviation from the guideline should be well thought out and documented by the professionals, which may be influenced by available information for decision making, professionals’ choice or patient preferences. Unfortunately, we were not able to include arguments to deliberately deviate from guideline recommendations, since these are frequently not documented (in a standardized way) in medical records.

Previous studies found patients’ age as an important factor for delivered NHL care: they studied elderly DLBCL patients, defined as patients aged over 60 or 75 years [11, 12, 25]. Younger age and better performance status were associated with receiving CHOP-like chemotherapy. Van de Schans et al. [12] showed age as the only factor associated with receiving less than six cycles of CHOP-like chemotherapy (adjusted for variables as gender and co-morbidity). Concerning overall survival, all three studies concluded that optimal therapy for elderly was associated with better outcomes, after case-mix corrections [11, 12, 25]. After multivariate analyses, Trebouet et al. [35] found also a relation between treatment administration and improved survival in patients over 90 years of age with aggressive NHL. An important drawback of intensive chemotherapy is treatment related toxicity. The elderly are more susceptible to complications, which makes it even more important to accurately select patients for therapy [11]. They stated that elderly are more susceptible to develop complications, which makes it even more important to accurately select patients for therapy. The judgment of professionals must be underscored in this selection process. A possible option to optimize outcomes was proposed by Lin et al. [25]; they opted implementation of tailored interventions to improve the performance status of patients before the start of therapy. In addition, in other fields of oncology lower guideline adherence was seen for elderly as well [34, 36]. Suggested reasons for the lower rates were that elderly patients receive less diagnostics and/or therapy for medical reasons, such as higher burden of co-morbidities [34], or diagnosis of advanced disease stages [36], which was initially seen in our dataset as well (data not explicitly shown). However, co-morbidities and disease stage were included in our analyses and age remained a determinant in the final models.

Besides age, several other tumor and patient related determinants were involved in explaining hospital variation, including previous malignancies, LDH and Hb level, gender, co-morbidity, extranodal involvement, tumor type and tumor aggressiveness. Most of these aspects are common factors measured in NHL research concerning prognostic factors and survival analyses [11, 12, 25, 35]. Unfortunately, this literature shows involvement of the factors with survival in univariate analyses, but not in multivariate analyses. Tumor type and aggressiveness are often not assessed, since studies regularly select only DLBCL or aggressive tumors as subjects of interest [11, 25, 37]. Kuper-Hommel et al. [37, 38] investigated differences in therapy and outcome between patients with nodal and extranodal lymphomas in two large population-based studies. They showed that patients with extranodal lymphomas were less often optimally treated but did not find clear differences in overall survival. In our study, patients with extranodal involvement received less often all required staging techniques and showed more often dose reductions during R-CHOP chemotherapy or reductions without reporting the reason.

Not all determinants found seem directly relevant for clinical practice, such as the influence of the Hb level on QI18: pathology results have to be known before the start of treatment. A possible explanation could be that the urge of starting therapy is higher for patients with a aberrant Hb level and an aggressive tumor. It seems valuable to explore these determinants in other NHL populations.

Of the professional and hospital related determinants for hospital variation in NHL care, treatment is an important factor in relation to better survival, as discussed above. Factors as MTC, hospital region, in-hospital referral, PET-scanner and discussion in a pathology panel are often not taken into account in survival analyses. The possible relation of these factors with overall survival is an interesting issue to address in future research. Hospital region will probably be one of the most challenging determinants, since hospitals cannot move to another geographical region and regional collaborations are embedded, which might be tough to effect change upon. Nevertheless, guideline adherence and quality of care described per region can give valuable insight into regional differences concerning interpretation and rating of the guideline recommendations and provide possible points of interest for improving quality of care.

Strengths of this study are the large study sample (N = 423) derived from a population-based cancer registry and the validated guideline-based QIs used for the assessment of variation in guideline adherence for NHL care. These factors contribute to the reliability of our results. Another factor contributing to a reliable dataset is that trained registration employees of the IKNL collected the data independently of the project team. An additional strength of our study is that 2 levels of potential determinants were included, namely patient and hospital level. Multilevel analyses made it possible to include these factors in one regression model per quality indicator.

There are also some limitations that need to be addressed. First, characteristics at the level of professionals were not taken into account, since NHL care is provided by a multidisciplinary team of a hematologist and/or (radiation)oncologist, radiologist, nuclear physician, pathologist and oncology nurse. It was not possible to relate one professional to one patient, which is necessary for inclusion of characteristics at professional level. However, some professional related factors measured at patients level were included in our study, such as patients discussed in MTC and therapy used. Second, only two of eight hospital characteristics included for analyses were found to have significant impact in the final multilevel models. This can be caused by the limited sample size of 19 hospitals, indicating more hospitals may be needed for possible future research. Third, no hospitals from the Western part of the Netherlands were included in our study, which might have introduced some selection bias. However, we did include 19 of the 91 Dutch hospitals, including three different regions, representing 21 % of the Dutch hospital population. Last, a significant amount (>50 %) of data was missing for the parameters performance status and IPI score. One of the reasons for this could be that only official WHO scores and Karnofsky scores were collected, excluding general terms as ‘healthy man’ or ‘vital women’. Arguments for not calculating the IPI score included that therapy choices do not change for most patients based on the IPI score, except for patients participating in clinical trials.