DISCUSSION
This national retrospective demonstrated that stage, the use of chemotherapy and an elevated serum AFP level at diagnosis were independently predictive of overall survival. Additionally, female sex was associated with a significantly lower risk of death (HR 0.284; p-value 0.037). The MaGIC (Malignant Germ Cell Tumour International Consortium) Study of 2015 revealed that stage, age and tumour site were significant predictors of outcome but not sex (24). A recent epidemiological study of paediatric germ cell tumours conducted in Danish children revealed a similar female predominance (71%) but did not show any relationship between sex and risk of death (33). Neither histology nor nutrition were predictive of outcome in our cohort, although in the MaGIC study yolk sac tumours were reported to have a better outcome, although not found to be statistically significant (24).
In our study tumour stage was significantly predictive of outcome (p<0.001), especially for patients with advanced disease (Stage III (p=0.0171) and IV (p<0.001)). Our findings therefore correlated with the MaGIC analysis which demonstrated a statistically significant relationship between stage IV disease, and poor outcome, particularly in children over 11 years of age with advanced stage extragonadal tumours and metastatic ovarian tumours (24). Earlier studies also documented that advanced disease was predictive of poorer outcomes (16, 34).
In our cohort a serum AFP level at diagnosis of >33,000ng/ml was associated with a poorer OS and a level of > 21,000ng/ml associated with a poorer EFS (p = 0.044). Previously, serum AFP levels of more than 100,000μg/L have been associated poorer outcomes in children with hepatoblastoma (35). The MaGIC analysis found that an elevated serum AFP level >10,000 was associated with a poorer outcome but it did not reach the definition for statistical significance (24). Alpha fetoprotein as a tumour marker has been described as tumour-associated rather than tumour-specific, given that its production can be elevated in a variety of both benign and malignant conditions in children and that the persistence of elevated levels acquired in utero, which usually normalises by 10 months of age ex utero, can sometimes persist until 2 years of age. The elevation of serum AFP levels cannot therefore be considered diagnostic on its own and must be interpreted in concert with other factors (36).
The MaGIC study was able to demonstrate a threshold of age (11 years) beyond which outcomes were poorer using a bootstrap technique (24). We interrogated the same question using age modelled as a penalised spline because of our smaller sample size, and limiting it to patients with more advanced disease, but could not demonstrate a threshold of age beyond which outcomes were either more or less favourable.
In addition to female sex, higher household incomes, in excess of USD 23000 per year, was associated with a significantly lower risk of death (HR0.071; p=0.039). We also showed a significant relationship between lower stage disease and higher SES. (p=0.03). However, early presentation alone could not account for this association. This suggests that there are SES factors which may impact outcome during treatment. For example, access to transport may cause delays in chemotherapy or food insecurity may lead to malnutrition, negatively impacting immunity and the ability to combat infection or tolerate chemotherapy-related toxicity. Given our past of enforced racial segregation and the association between skin colour and poverty, it is critical to isolate these social determinants of health as skin colour is too easily invoked as a risk factor, when in reality these relate to privilege and access to care. In a large (n=3173) population-based study in adult men with advanced stage prostate cancer in the US, in which ‘race’/ethnicity was self-declared, socio-economic status together with clinical and pathological factors accounted for a 15% increased relative risk which was independently predictive of outcome in men of African-American decent. The relationship, however, did hold true for Hispanic patients on multivariate analysis demonstrating the sometimes tenuous relationships between these factors and the necessity for the identification of clear risk factors for outcome (31). In children with acute lymphoblastic leukaemia a single institutional study reported a higher risk of relapse in children from high poverty areas despite uniform treatment (32). This is relevant given our intention to embark on the implementation of a national treatment protocol for MEGCTs and will be instructive as we interrogate household income and maternal education as measures of SES. Also proving a significant relationship between SES and outcome strengthens our collective call for parity not only in service delivery and access to care but also for adequate social support for children, which is constitutionally enshrined, but not necessarily delivered in real time.
Analysis of our treatment interventions demonstrated that the surgical local control rate in our study was high with a complete resection rate 64.8%. Although the numbers were small it is a notable that patients who did not undergo re-exploration for negative imaging at first assessment, and who therefore did not undergo second look surgery when their imaging revealed no demonstrable disease, died. The critical role of local control (surgery in the case of MEGCTs) in paediatric solid tumours is incontrovertible and cannot be under-emphasized (37). Additionally, there was a 5% surgical refusal rate which is a cause for concern considering that half of those patients died of disease as a result. In a systematic review of factors influencing refusal of therapy in adult cancer patients, discomfort of treatment, fear of side effects and transportation difficulties were cited as the most important contributing to refusal of surgery (38). In a recent study from Vietnam adherence to cancer treatment in children was found to be determined by prognosis whilst being associated with local perceptions of cancer and the economic power of affected families (39). It is possible that a more rigorous informed consent and assent process such as that proposed for a national treatment protocol may help to allay the fears associated with paediatric cancer treatment specifically with respect to the surgery which remains a cornerstone of care.
By comparison, the use of chemotherapy was strongly predictive of outcome with no single regimen performing better than any other (carboplatin vs. cisplatin-based) and all regimens producing superior results compared to no chemotherapy at all (p<0.001). This is corroborated by a recent MaGIC report demonstrating equipoise in contemporaneous trials using cisplatin compared to carboplatin-based regimens in paediatric germ cell tumours (19).
Radioisotope GFR measurement is recommended as part of the standard of care guidelines for renal function assessment in paediatric oncology practice as it impacts the dosing of platinum alkylators. As such standardising methodologies is critical (40). Rates of formal GFR assessment (even in the absence of radio-isotope instruments) and also audiology assessment were infrequent, which is related to the lack of availability of or access to these services at all POUs across the country, as well as the large burden of patients presenting with advanced disease, whose poor performance status (renal or pulmonary) may preclude radioisotope testing at the time diagnosis. Similarly, the use of calculated GFRs was low in the absence of formal radio-isotope testing. In a more recent study calculated GFRs were found to either over or underestimate the GFR. The recommendation was that radioisotope GFR determination should remain standard of care for platin dosing to ensure accuracy (41). Unfortunately, this may not always be possible in our setting and a calculated GFRs may be the best method to determine glomerular function when no radio-isotope determination is available. The value of calculated versus radioisotope GFR measurement can be prospectively tested in the context of a national protocol and calculated GFRs as a surrogate for more accurate renal function evaluation could become standardised practice.
In addition, to renal function assessment, significant numbers of hearing-impaired children may have evaded detection because of low rates of routine audiology testing which has the potential to impact school performance and subsequently employability. This is notable because the unemployment rate for the hearing impaired in South Africa ranges from 21.2% - 35.3% with low average annual household incomes of approximately 2000 USD for the moderately hearing impaired and 1500 USD for those with severe deficits (42), placing them close to the upper limit poverty line of 800USD per household per year (43).
Treatment-related mortality was low. Three patients died from infections, two from chemotherapy toxicity and one from surgical complications. Although a small number of complications is likely to be inevitable, it would be instructive to determine whether the rate of complications can be reduced with the implementation of a standardised chemotherapy and surgical guidelines. Despite this, the 5-year OS for the entire cohort was very encouraging (80.4%) and patients who relapsed were salvaged (5-year EFS 75.3%).
We recognise that the sample size and hence the number of events were small and that this could have limited the statistical power to detect significant effects. Different treating centres made use of institutionally specific treatment protocols, often determined by resource constraints. Where space and bed capacity were at a premium, outpatient approaches, like JEb was favoured over inpatient regimens like BEP or PEb. Access to formal radio-isotope testing and audiology testing was fragmented across POUs, limiting the accurate assessment of renal function for platin dosing and hearing assessments.
This retrospective analysis was undertaken to assist with the development of a national protocol for children with MEGCTs in South Africa to standardise management and toxicity assessments. In the new national protocol the choice of platinum will be influenced by several factors: firstly, previous local reports of the cost effectiveness of JEb as an outpatient strategy (22) and the significant relationship between SES and outcome reported here; secondly, the rates of platin-related ototoxicity reported here and that under-testing has potentially resulted in a failure to detect significant numbers of hearing impaired children, and; lastly, the availability of audiology services and nuclear medicine facilities which are not ubiquitously available and tend to be clustered near larger POUs in wealthier urban areas.
In addition, the finding of significant relationships between serum AFP levels and SES opens opportunities for the interrogation of these factors in a prospective way in the context of the new protocol as we seek to provide guidance for best practice and as we advocate for social justice in medicine. Creating the discipline of standardised practice by teams with disease-specific expertise able to collaborate across institutional divides will improve paediatric oncology care in South Africa. Thereafter, we can advocate for the establishment of bilateral funding agreements, to open doors to children and families from across the sub-continent to receive expert, coordinated care at our national centres of excellence.