Development of an oncological-multidimensional prognostic index (Onco-MPI) for mortality prediction in older cancer patients

In the present study, the cancer-specific Onco-MPI appeared to be a highly accurate and well-calibrated prognostic tool for one-year mortality in older cancer patients that can be useful for defining homogeneous prognostic categories and clinical decision making in this age group. Indeed, therapeutic decisions in elderly cancer patients are not fully informed unless heterogeneity of the aging process is taken into account. Actually, some forms of CGA have been successfully used to establish individualized treatment plans of treatment (Caillet et al. 2011) and in defining risk of toxicity from treatments in older cancer patients (Hurria et al. 2011; Extermann et al. 2012). Thus, current clinical guidelines for cancer in older age recommend to implement the CGA methodology (Extermann et al. 2005; Biganzoli et al. 2012; Droz et al. 2010; Pallis et al. 2010) in order to determine the residual biological, psychological and functional capabilities of the older patients, i.e., the grade of frailty (Baijal and Periyakoil 2014; Hamaker et al. 2012), for developing a personalized plan for treatments and interventions. Indeed, whatever the definition and methodology used to evaluate frailty, frail patients have a higher mortality compared to non-frail patients. In a previous study, we showed that frail hospitalized patients, being treated despite poor conditions, had poor outcome (Basso et al. 2008). Furthermore, frail lymphoma patients had same outcome whether they were treated with active modified oncological treatment or palliative care (Tucci et al. 2009). The prognostic evaluation of life expectancy emerges thus as a key factor by which pros and cons of active oncological treatment must be weighted, both in the adjuvant setting and in the metastatic setting. Prognosis could be also fundamental for balancing the harm–benefit and cost–benefit ratios in situations of uncertainty when prescribing high-cost drugs or treatments requiring multiple admissions with potential impact on quality of life. In recent years, some prognostic scores have been proposed, but none of these was based on information collected by a standardized CGA (Yourman et al. 2012; Pilotto et al. 2015; Baijal and Periyakoil 2014). The MPI has been previously validated in older hospitalized patients suffering from major diseases, including several types of cancer (Pilotto et al. 2012a, b, c; Giantin et al. 2013), with a significant higher predictive power than other widely used frailty indexes (Bellera et al. 2012). In this cohort from an outpatient setting, we developed a modified MPI that included, compared to the originally MPI, the ECOG performance status (instead of the Exton-Smith scale), the MMSE (instead of the Short Portable Mental Status Questionnaire, SPMSQ) and the BMI (instead of the Mini Nutritional Assessment, MNA) to evaluate functional, cognitive and nutritional status, respectively.

Notably, we further included in the prognostic model both patients’ cancer site and cancer stage variables, along with the main MPI domains, due to their clinical relevance on the mortality risk prediction. For tumor sites, some biases can have been introduced since breast cancer was the prevalent type (46.5 %) with early stages of disease being more represented for this tumor site, whereas other tumor sites, i.e., lung cancer, were far less prevalent, with higher stages of disease at first access. Globally, 75 % of patients received active oncology treatment whose impact on survival was not significant. Therefore, considering multidimensional aggregate information may go beyond the heterogeneity of the sample related to diverse cancer sites and stages, and the variety of treatments used (endocrine agents, several types of chemotherapy regimens).

While screening tools based on abbreviated CGA showed high sensitivity with low specificity in predicting mortality (Smets et al. 2014; Bellera et al. 2012), CGA-based prognostic scores similar to the Onco-MPI in older cancer patients have been poorly investigated. Recently, a study conducted on 249 older Asian cancer patients by using CGA items to build a prognostic nomogram based on six clinical laboratory items demonstrated a relatively low predictive accuracy for one-, two- and three-year overall survival with a C-index value of 0.71 (Kanesvaran et al. 2011). The Onco-MPI was developed on more than 600 patients using some CGA-based items which can be easily implemented in routine oncological practice to drive treatment decisions. A time horizon of 1 year was chosen because we focused on a short-term mortality risk prediction, which is particularly helpful to give systemic oncological treatment or not in many cancers of the older people.

In this study, ECOG performance status was significantly associated with mortality, thus confirming the well-known prognostic role of this tool in oncology. Moreover, both ADL and IADL in our model were significantly associated with mortality in univariable analysis, and the loss of significance in the multivariable analysis probably was related to interaction with other considered domains (Extermann et al. 1998). The presence and number of severe comorbidities have also been found to be related to poor prognosis in the present study. Indeed, comorbidity has been consistently proven to be associated with worse survival in older cancer patients (Piccirillo et al. 2004). This has been shown to be independent of ECOG performance status (Firat et al. 2002) and functional status (Extermann et al. 1998). Weight loss or a low BMI was associated by reverse epidemiology with an increased risk of mortality in older age (Landi et al. 2000), a finding confirmed also in our cohort of cancer patients Furthermore, cognitive status showed a strong prognostic value in our older cancer patients, confirming earlier findings suggesting dementia as an independent prognostic factor for survival in older subjects (Wolfson et al. 2001). Depression has been also proposed as a predictive factor of mortality in older cancer patients with controversial findings (Satin et al. 2009). In our study, the presence of depression was not associated with survival, and therefore it was not considered among variables for building the Onco-MPI. In both the geriatric and oncology literature, social isolation has been linked to an increased risk of mortality (Kroenke et al. 2006). In our model, the presence of a caregiver was unexpectedly related to worse survival, probably because in an outpatient setting, those requiring a caregiver could have been the most vulnerable ones.

The calculation of Onco-MPI can be easily performed through an excel file in which the value of each domain (age, sex, BMI, ADL, IADL, ECOG PS, MMSE, number of severe comorbidity, cancer stage, tumor site, number of drugs, the presence of caregiver) is multiplied by the coefficient which is the weight for Onco-MPI (reported in Table 3).

From a practical standpoint, we can take the example of an 80-year-old woman with stage III colorectal cancer, PS 1, ADL 6/6, IADL 5/8, BMI 28, good cognitive status (MMSE ≥ 24), two severe comorbidity, five drugs, with caregiver present; this patient has an Onco-MPI score of 0.44, which means that she is in the lower Onco-MPI score, with an estimated risk of mortality at 1 year of 2.1 %. The same woman, but having a bad cognitive status (MMSE < 24) and BMI 19, has an Onco-MPI score of 0.54, which means that she is in the intermediate risk group, with an estimated one-year mortality of 17.7 %. If this patient had ADL 5/6, her Onco-MPI would be 0.72, which corresponds to higher risk of mortality at 1 year, estimated to be more than 80 %. While adjuvant treatment may be discussed in the first case and carefully evaluated in the second case, in the third case geriatric assessment and oncoMPI suggest not to consider adjuvant chemotherapy. It is important to notice that the patient was judged as having a PS 1 in all cases, thus confirming that comprehensive geriatric assessment adds information, and in the case of the Onco-MPI adds prognostic information, to simple PS estimation (Repetto et al. 2002).

The main limitations of the present study were the possible selection of more “fit” older cancer patients and the lack of data of the predictive role of the Onco-MPI for mid- and long-term mortality. The Onco-MPI has been built on outpatients; therefore, there is a greater likelihood of the inclusion of less frail patients which may have skewed the results. However, it holds a very good discriminatory power with a C-statistic of 0.87. In fact, despite the lack of accepted criteria to assess the quality of prognostic indices, generally C-statistics for discrimination can be considered good for ranges 0.70–0.79, very good for ranges 0.80–0.89, and excellent for 0.90 or greater (Yourman et al. 2012). Beyond its prognostic ability, the Onco-MPI could also serve as an useful tool for evaluating effectiveness of an intervention in different settings, when changes in Onco-MPI categories that can be obtained may reflect the outcome of the intervention, as recently reported in older patients with late-life major depressive disorder responders and non-responders to antidepressant treatment (Pilotto et al. 2012a, b, c). Finally, the Onco-MPI in the research setting may help to properly classify patients enrolled in clinical trials, selecting more homogeneous subgroups of patients. However, the Onco-MPI warrants external validation, which is already underway, and proof of ability to predict mortality with longer follow-up.