Skip to content

Ipi Research Paper

The current wave of technological change has created new opportunities for multilateral cooperation across a wide range of areas, including sustainable development, conflict prevention, humanitarian responses, peace operations, and state-society relations. At the same time, however, it has created an enduring “digital divide,” raised questions about Internet governance and privacy, and led to new forms of warfare that challenge existing international human rights and humanitarian laws.

The UN has at times struggled to keep up with the pace of change, in part because private sector and civil society actors are often in the lead when it comes to technological innovation. This policy paper explores where the UN can play a useful role and where existing mechanisms and other actors are better placed. Based on extensive consultations with representatives of states, various UN entities, and civil society, as well as subject-matter experts, this paper details recommendations laid out in the ICM’s final report, published in September 2016. These include to:

  • Consolidate a multilateral space for innovation and new technology; and
  • Recognize the Internet and big data as global public goods.

To stand with those who are committed to working multilaterally and reforming the international community, we are asking people to use the hashtag #MultilateralismMatters. For more, including sample tweets and graphics, read IPI’s Social Media Toolkit here.

Diffuse large B cell lymphoma (DLBCL) is the most common histological subtype of all Non-Hodgkin Lymphomas (NHL) and comprises about 20% of newly diagnosed lymphoid neoplasms (Morton et al, 2006). The international prognostic index (IPI), based on clinical parameters, is a common tool used for prognostication and is also valid in the rituximab era (Ziepert et al, 2010). Nevertheless, despite the establishment of a revised IPI in the rituximab era (Sehn et al, 2007; Zhou et al, 2014), its prognostic value is suboptimal in high risk DLBCL patients. Furthermore, prognostic tools based on new molecular insights in the pathobiology of DLBCL are currently far from routine clinical application and thus there is a relevant clinical need for more accurate prognostication in DLBCL patients.

Recently, the National Comprehensive Cancer Network (NCCN) proposed an enhanced IPI (NCCN-IPI) in patients treated with R-CHOP (rituximab, cyclophosphamide, doxorubicin, vincristine and prednisone) at seven NCCN-cancer centres in the United States, which was validated in the British Columbia Cancer Agency (BCCA) (Zhou et al, 2014). This score emphasizes the impact of older age and very high lactate dehydrogenase (LDH) levels and clearly defines extranodal disease relevant for prognosis. However, the value of the NCCN-IPI in patients outside North America remains to be established and important laboratory parameters, except LDH, have not been collected. A possible selection bias, especially in the population treated at the NCCN cancer centres in the USA, was also discussed (Zhou et al, 2014) and seems likely, considering the young mean age and good performance score of the included patients [NCCN cancer centres: mean age of 57 years, 89% of patients had an Eastern Cooperative Oncology Group (ECOG) performance score of 0–I].

We thus aimed to validate the NCCN-IPI in a presumably less selected population of European DLBCL patients and analyse the impact of additional laboratory parameters on the NCCN-IPI.

Patients and methods

Patients included for this analysis were diagnosed with DLBCL between 2004 and 2013. All patients were negative for human immunodeficiency virus and treated with R-CHOP or R-CHOP-like chemoimmunotherapy as first line treatment at the 3rd Medical Department of the Paracelsus Medical University Salzburg, Austria and the Division of Haematology of the Medical University of Graz, Austria. Clinical data, including the ECOG performance score, B-symptoms, stage according to Ann Arbour, overall survival (OS) and progression-free survival (PFS), were retrospectively analysed by chart based review. These clinical data were used to centrally review and assess the IPI and NCCN-IPI. All laboratory data were obtained before start of treatment as per clinical routine. For patients who did not attend to follow-up visits, clinical follow-up data were obtained by telephone interviews with the patients' general practitioner.

This retrospective analysis was approved by the Ethics Commitee of the provincial government of Salzburg, Austria (415-EP/73/127-2012) and the local ethical committee of the Medical University of Graz (No. 25-434 ex 12/13). Statistical analyses were performed using IBM® SPSS® statistics software, version 21 (SPSS Inc. Chicago, IL, USA) and R-statistical software environment (including the survival and CPE packages) ( Mann–Whitney-U-test and Pearson's chi-squared test were used for univariate analyses, where appropriate. Survival was estimated using Kaplan–Meier curve analysis, with statistical comparisons using the log-rank statistic. A two-tailed significance-level of 0·05 was considered statistically significant. Only statistically significant factors were included in the multivariate Cox-regression analyses. Discrimination of the different scores was assessed using the concordance probability estimate (CPE) with a 95% confidence interval (CI) for OS (Gönen & Heller, 2005). Inter-raterweighted kappa statistics were used to compare the degree of agreement in risk stratification.


Patient characteristics

A total of 499 consecutively diagnosed DLBCL patients treated with first line R-CHOP or R-CHOP-like treatment at two academic cancer centres in Austria between 2004 and 2013 were included in this retrospective analysis (Table 1). In detail, advanced disease according to Ann Arbour was present in 47·3% of the patients and 52·7% were male. Non-pegylated liposomal doxorubicin (NPLD) was used in 25·5% of patients. The mean age of the whole cohort was 65·3 years (range 20–92 years), which is higher than in the population used for establishment of the NCCN-IPI (mean age: 57 years) (Zhou et al, 2014). There was no significant difference in age, stage according to Ann Arbour, ECOG performance score, frequency of extranodal involvement or distribution of the NCCN-IPI between patients treated in both participating Austrian cancer centres (Table 1). The median follow-up for all patients alive was 51 months.

Age (years)
Mean ± SD65·3 ± 14·666·9 ± 1463·9 ± 14·80·646a
>60 years (%)70·776·766·1
Sex (%)
Stage (%)
Low (1–2)52·851·153·90·53b
High (3–4)47·248·946·1
IPI (%)
Low (0–1)34·528·838·90·03b
Low-intermediate (2)30·333·827·5
High-intermediate (3)22·821·923·6
High (4–5)12·415·510·0
Low (0–1)9·26·411·40·1b
Low-intermediate (2–3)39·939·340·4
High-intermediate (4–5)37·938·437·5
High ≥613·016·010·7
B-Symptoms (%)
Extranodal Involvement (%)
Increased LDH (%)
LDH-R ≤ 140·133·345·40·02b
LDH-R > 1–350·155·745·7
LDH-R > 39·811·08·9
Albumin (%)
<35 g/l24·330·520·50·02b
Available (n/total)403/499154/219249/280
β2-microglobulin (%)
>3·0 mg/l4648·343·20·41b
Available (n/total)263/499145/219118/280
Liposomal doxorubicin (%)

Prognostic impact of the NCCN-IPI

Applying the NCCN-IPI in our patient cohort resulted in 9·2% of patients being classified as low risk, 39·9% as low-intermediate, 37·9% as high-intermediate and 13% as high risk. In comparison to the conventional IPI, fewer patients were classified as low risk (Table 1). As expected, the NCCN-IPI identified four prognostic groups of patients with highly different PFS (P < 0·001, Fig 1a) and OS (P < 0·001, Fig 1b). Furthermore, the NCCN-IPI was more accurate in predicting 3- and 5-year PFS and OS in comparison to the conventional IPI (Table 2).


Modified scoring of the NCCN-IPI in the elderly

The mean age at initial diagnosis of DLBCL is about 65 years in the Caucasian population and therefore the majority of patients comprises the elderly (Shenoy et al, 2011). The new NCCN-IPI attributes more prognostic impact to higher age, excluding patients >60 years from the low risk category (0–1 score points). Therefore, we hypothesized that the ability to differentiate risk groups in elderly patients might be impaired.

Analysing only the patients older than 60 years, the three applicable NCCN-IPI risk groups: low-intermediate group (2–3 score points), high-intermediate group (4–5 score points) and high-risk group (≥6 score points) also discriminated three different prognostic groups but with lower differences between the patients classified as low-intermediate or high-intermediate (3-year OS: 81·5% vs. 66·5% and 5-year OS: 71·1% vs. 54·1% P < 0·001) than between high-intermediate or high risk patients (Table 3).

Low (0–1 points)n.a. in elderlyn.a. in elderlyn.a. in elderlyn.a. in elderly
Low-intermediate (2–3 points)71·881·563·471·1
High-intermediate (4–5 points)62·666·550·954·1
High (≥6)38·639·530·932·3
Alternative NCCN-IPI scoring for elderly
Low (2 points)83·494·275·489·0
Intermediate (3–5 points)63·569·352·556·5
High (≥6)38·639·530·932·3

We thus propose that the low-risk elderly group (12% of patients in our cohort) be defined as having exactly 2 NCCN-IPI score points, allocated for age >60 years; the intermediate group (69·6% of our cohort) should be defined as 3–5 score points and that the original NCCN-IPI definition of high risk should be maintained as ≥6 points (18·4% of patients in our cohort). This strategy identifies a group of elderly low risk patients with excellent prognosis (3-year PFS/OS: 83·4/94·2% and 5-year PFS/OS: 75·4/89%), a group with intermediate prognosis (3-year PFS/OS: 63·5/69·3% and 5-year PFS/OS: 52·5/56·5%), whereas the high risk group remains unchanged (3-year PFS/OS: 38·6/39·5% and 5-year PFS/OS: 30·9/32·3%; P < 0·001) (Table 3, Fig 2).

Prognostic role of albumin and β2-microglobulin in the NCCN-IPI

Although the NCCN-IPI incorporates the negative prognostic impact of very high LDH serum levels, other laboratory parameters have not yet been evaluated (Zhou et al, 2014). The pretreatment serum levels of albumin and β2-microglobulin were available in 80·8% and 52·7% of the 499 patients included in our study. Albumin level <35 g/l and β2-microglobulin >3·0 mg/l were independent prognostic factors for a worse OS in univariate and multivariate Cox regression analyses [Albumin hazard ratio (HR): 1·97, 95% CI: 1·12–3·47 P = 0·018; β2-microglobulin HR: 2·16, 95% CI: 1·16–3·99 P = 0·014; Table 4].

Male vs. female0·8330·602–1·1510·268n.a.  
Low vs. high3·0682·146–4·386<0·0012·3081·194–4·4600·013
β2-microglobulin >3·0 mg/l
Low vs. high3·2872·045–5·286<0·0012·1611·168–3·9980·014
Albumin <35 g/l
No vs. yes3·1982·175–4·703<0·0011·9781·126–3·4750·018

To exclude a selection bias in the patients with available albumin and β2-microglobulin we tested for a significant difference of distribution of NCCN-stages and age in comparison to the patients without these parameters. There was no difference in the distribution of all NCCN-risk groups (P = 0·974) and in mean age of patients (65·6 vs. 64·9 years P = 0·698).

Due to the wide availability of serum albumin in our cohort and its high significance in multivariate analysis (Table 4), we incorporated this parameter in a risk prediction, based on the NCCN-IPI. Low serum albumin was allocated as a risk factor with two points because the risk ratio was similar to the overall effect of the NCCN-IPI in multivariate analysis. In addition to the NCCN-IPI, this resulted in a new score with a maximum of 10 points. Using this score, a large group of patients at low risk (0–2 score points) with excellent OS (3-year OS: 97·8% and 5-year OS: 93·5% n = 99), an intermediate-low risk group (3 score points; 3-year OS: 82·7% and 5-year OS: 78·0% n = 75), an intermediate-high risk group (4–7 score points; 3-year OS: 65·9% and 5-year OS: 55·7% n = 198) and a high risk group (8–10 score points; 3-year OS: 44·2% and 5-year OS: 36·8% n = 31; P < 0·001) could be identified. This strategy widens the definition of low risk patients, but maintains their excellent prognosis in comparison to the original NCCN-IPI in patients with available serum albumin and maintains the accuracy in higher risk patients (n = 403; low-risk: 3-year OS: 97·0% and 5-year OS: 92·4% n = 36; low-intermediate: 3-year OS: 88·5% and 5-year OS: 82·6% n = 160; high-intermediate: 3-year OS: 68·1% and 5-year OS: 59·3% n = 156; high-risk: 3-year OS: 40·3% and 5-year OS: 28·8% n = 51; P < 0·001) (Fig 3).

Superior discrimination of the new prognostic scores

As expected, the NCCN-IPI outperformed the IPI (CPE: 0·753 vs. 0·713) with a higher CPE in discrimination regarding OS in all patients, but the modified NCCN-IPI including the value of serum albumin was even superior to the conventional NCCN-IPI (CPE: 0·783 vs. 0·753) in our cohort. This effect was similar when the CPE was tested only in the elderly (CPE: IPI 0·637 vs. NCCN-IPI 0·674 vs. modified NCCN-IPI 0·723). Weighted kappa statistics indicated a high level of agreement of risk prediction by the conventional NCCN-IPI and the modified NCCN-IPI (weighted kappa: overall cohort – 0·83, elderly subset – 0·82).


After 20 years of use for risk prognostication in DLBCL patients, the standard of the conventional IPI seems to be challenged by the new NCCN-IPI (Zhou et al, 2014). The NCCN-IPI risk score differentiates patient age into four risk groups and includes the negative impact of very high LDH levels. Furthermore, it focuses on the involvement of defined high-risk extranodal sites and not on the number of involved extranodal sites leading to a more precise definition. Nevertheless, validation outside North America, information about the use of the NCCN-IPI in elderly patients and other laboratory parameters other than LDH are currently lacking.

The present study confirms the utility of the NCCN-IPI in DLBCL in European patients. Despite the higher age and more patients with increased LDH than in the original population of the NCCN-IPI (Zhou et al, 2014) we showed that the NCCN-IPI is indeed a powerful tool to predict PFS and OS in 499 DLBCL patients treated at two different Austrian Academic cancer centres. The NCCN-IPI also more accurately identified patients as low or high risk (see Table 2). Both participating cancer centres are responsible for the care of almost all lymphoma patients in their district and this type of treatment is not offered in smaller hospitals in their region. In addition, there are no haematologists and oncologists in private practise in Austria. We thus believe that our cohort is most likely much less selected that patients treated at NCCN cancer centres and is thus relatively close to the ‘real life setting’ of DLBCL patients. These circumstances and the higher age of our patients compared to the NCCN-cancer centre and the BCCA cohort may be responsible for the differences in the 5-year OS and the proportions of high and low risk patients between our cohort and the original report (Zhou et al, 2014).

Several other prognostic tools based on the cell of origin (COO) or other molecular features have been published (Meyer et al, 2011). Nevertheless, because of the lack of reproducibility of immunohistochemistry to assess COO (Gutierrez-Garcia et al, 2011) this cannot be considered as standard in clinical practise and risk prediction based on clinical parameters, such as the NCCN-IPI, is still largely the standard of care. Furthermore, this risk score should be assessed in future clinical trials and may be more useful than the conventional IPI to identify high-risk patients for more intensive treatment.

Non-pegylated liposomal doxorubicin was used in 25·5% of patients because of a lower rate of cardiac side effects shown in patients treated for breast cancer (Aapro et al, 2011). Therefore, NPLD was preferentially used in older DLBCL patients compared to conventional anthracyclines (75·2 vs. 61·8 years P < 0·001) in our cohort. NPLD has been used in a randomized trial of the Austrian Study Group of Medical Tumour Therapy (AGMT) and preliminary data in abstract form of this trial and a retrospective analysis of our own experience in the elderly show equal efficacy in DLBCL compared to conventional anthracyclines (Fridrik et al, 2011; Melchardt et al, 2014). Therefore, NPLD seems a rational choice especially in patients at high risk for cardiac toxicity.

Due to the higher impact of age in the NCCN-IPI, elderly patients are excluded from the low risk definition. Therefore, we hypothesized those elderly patients with only low-risk disease may be misclassified. We were able to show that elderly patients with 2-risk points due to age >60 years have a very favourable prognosis (3-year OS: 94·2% and 5-year OS: 89%) and should not be classified in the low-intermediate group together with patients with three risk points, as proposed by the NCCN-IPI (Table 3; Fig 2). These low risk patients with two risk points comprise fit patients older than 60 years that are not adequately captured by the NCCN-IPI risk score. The NCCN-IPI uses calendar age as a risk variable. It should be mentioned, as a note of caution, that fit patients in our cohort aged between 60 and 75 years with favourable biological risk factors were thus classified into a more serious risk group than necessary by the NCCN-IPI. In fact, our cohort includes a quite large population of patients older than 60 years without relevant comorbidities who do as well as the younger patients, if no other risk factors are present, but would be eligible for intensive treatment according to the NCCN-IPI.

The management of this population of ‘fit elderly’ is a common problem in clinical practise nowadays, because most treatment concepts are based on the calendar age of the patients. Tools to evaluate comorbities or frailty, such as the Carlson Index, or the assessment of daily live activities may be helpful, but are not currently widely used in clinical routine.

In the first description of the NCCN-IPI, no further information about laboratory parameters, except LDH, were obtained in the NCCN-cancer centre or BCCAcohort. Albumin and β2-microglobulin are described in the literature as independent prognostic factors in aggressive lymphoma (Steward et al, 1984; Cowan et al, 1989; Johnson et al, 1993). We confirmed the independent negative prognostic impact of low serum albumin and high β2-microglobulin in multivariate analyses despite the improved performance of the NCCN-IPI (see Table 4). We also tested a new model incorporating low serum albumin into the NCCN-IPI, which opens the low-risk definition to more patients without comprising their favourable prognosis. This suggests that favourable biology of the tumour and good performance of the patients represented by high albumin may overcome classical risk factors, such as age (28 of 99 patients older than 60 years), increased LDH (28 of 99 patients) or advanced stage (12 of 99 patients) (Fig 3). These results are hypothesis generating and should now be further validated in other cohorts of DLBCL patients.

In summary, we confirm the excellent performance of the NCCN-IPI in a European cohort and show that it is prognostic, even in the elderly, in a presumably less pre-selected population than in the original report (Zhou et al, 2014). We propose that low risk elderly patients should be defined as having exactly 2 NCCN-IPI score points, allocated for age >60 years; the intermediate group (68% of patients in our cohort) should be defined as 3–5 score points and that the NCCN-IPI definition of high risk should be maintained as ≥6 points. Additionally, we report that albumin and β2-microglobulin are likely to add significant information to the NCCN-IPI. It would be extremely interesting to investigate this modified NCCN-IPI risk score in other large cohorts and we suggest including the NCCN-IPI in the stratification of future DLBCL trials.

Authors′ contributions

T.M, K.T. and A.E are primarily responsible for the manuscript. All authors were involved in the management of the patients, wrote and critically revised the manuscript.

Conflict of interest

The authors declare that they have no competing interests with regard to the content discussed in the manuscript. This work was supported by the Paracelsus Medical University (PMU Grant: E13/17/089-MEG).