This laboratory-based study was conducted at the National Health Laboratory Service (NHLS) haematology department, Tygerberg Hospital (TBH), Cape Town, South Africa. Tygerberg Hospital is a tertiary public sector hospital. During the study period, patients with AML 13 years and younger were managed by the Paediatric Oncology unit, and those ≥ 14 years old were managed by the Clinical Haematology unit.

During the study period, adult patients with AML who were fit for intensive chemotherapy, based on age, performance status and co-morbidities, were managed with ‘7+3’ chemotherapy induction using daunorubicin and cytarabine, and were consolidated with two to three cycles of high-dose cytarabine. Patients who were unfit for intensive chemotherapy received either low-dose cytarabine or hydroxyurea or were referred for palliative care. There was no access to hypomethylating agents or venetoclax. During the study period, most patients over 60 years were not considered for intensive chemotherapy. Low-risk APL patients were managed with all-trans retinoic acid (ATRA) and idarubicin according to the Programa Español de Tratamientos en Hematología (PETHEMA) protocol,²⁸ and high-risk patients received ATRA, daunorubicin and cytarabine, according to the European APL protocol.²⁹ There was no access to arsenic trioxide during the study period. The only polymerase chain reaction (PCR) assay for measurable residual disease (MRD) monitoring available during the study period was PML::RARA and BCR::ABL. There was no access to MRD flow cytometry to guide post-remission treatment. Access to bone marrow transplant was also limited.

During the study period, paediatric patients were treated according to the BFM-2004 standard risk and high-risk treatment protocols, with five blocks of intensive chemotherapy using cytarabine, anthracyclines and etoposide.³⁰ Acute promyelocytic leukaemia patients received ATRA in addition to the above chemotherapy, in accordance with the BFM-2004 protocol.

A laboratory-based cross-sectional retrospective study of newly diagnosed AML patients over 3 years, from 01 September 2015 to 31 August 2018, was performed.

All adult and paediatric patients with a new AML diagnosis confirmed by bone marrow aspirate and trephine (BMAT) performed at TBH during the study period were included. Patients were excluded if they did not have their diagnostic BMAT or their treatment at TBH.

Data were collected from the computer databases via the NHLS Central Data Warehouse (CDW). All BMATs that were performed and reported at TBH with a diagnosis of AML were identified. Acute myeloid leukaemia response assessment BMATs and relapsed AML were excluded. The newly diagnosed cases of AML were further assessed on the NHLS laboratory information system (LIS) for the full blood count (FBC) at diagnosis, diagnostic bone marrow findings, cytogenetics, PCR and HIV status. The only molecular studies that were available during the study period were NPM1 and FLT3 qualitative PCR. Polymerase chain reaction and cytogenetic results were used to classify the AML according to WHO-HAEM5 subtypes and according to ELN or BFM-2012 risk stratification,^2,16,23 except for APL, which was risk stratified according to the Sanz criteria.²⁶ In the absence of genetic data, AMLs were subtyped according to differentiation. Secondary AMLs were classified according to ELN risk stratification, except AML in Down syndrome, which was classified as favourable risk.^31,32,33 For survival data, the hospital electronic administrative system (Clinicom) was used.

Statistical analysis was performed on Microsoft Excel and STATA 19 (StatCorp LLC, Texas, US). Descriptive statistics were performed to assess the demographic information. Frequencies were calculated for categorical data. A Pearson’s Chi-squared test was used to assess categorical variables. Median and interquartile range (IQR) were calculated for continuous data. A Kruskal–Wallis rank test was performed to compare the ages between categorical variables. To assist with comparison to international literature, the paediatric age group was analysed using both the traditional age stratification of < 18 years old, and according to the local age distribution of patients treated by the Paediatric Oncology unit during the study period (≤ 13 years). Survival analysis was performed using Kaplan–Meier survival curves, with follow-up censored on 01 September 2021. Survival function distributions were compared with a log-rank test. Cox regression analysis with Breslow methods was performed to assess the significance of multiple risk factors on survival and to estimate hazard ratio (HR). A p-value of less than 0.05 was considered significant.

This study was approved by the Stellenbosch University Human Research Ethics Committee (S21/06/099). A waiver of informed consent was granted, and the study was conducted using retrospective, de-identified data in accordance with relevant regulations.

Karyotype results were available in 68% (61/90), while 22% (20/90) had failed karyotyping and 10% (9/90) had no karyotype requested. The median turn-around-time for karyotype results was 13 days (IQR 7–21). Fluorescence in situ hybridisation (FISH) was performed in 50% (45/90) of patients. Cytogenetically normal-acute myeloid leukaemia made up 20% (18/90) of cases. Nucleophosmin protein-1 gene and FLT3 PCR was performed in 10% (9/90) of patients and was not detected in all cases. Only 63% (57/90) of patients had sufficient cytogenetic and molecular testing for diagnostic subtype and prognostic classification. Despite local protocols recommending mutation testing for patients with CN-AML, only three patients of the 18 patients (17%) with CN-AML had testing.

The majority (88%, 79/90) of cases were de novo AML. Acute myeloid leukaemia with defining genetic abnormalities comprised 54% (49/90) of cases, with APL being the most common subtype (21%, 19/90) (Table 1). Acute myeloid leukaemia, myelodysplasia-related (AML-MR) was the second most common subtype, representing 19% (17/90) of cases; this included 13 de novo cases and four secondary to MDS (Table 1). The median age of patients with AML-MR was 63 (IQR 36–75). Acute myeloid leukaemia defined by differentiation comprised 38% (34/90) of AML cases. The secondary AML cases included AML secondary to cytotoxic therapy, AML with antecedent haematological disorders and AML with germline predisposition conditions. The latter occurred in patients with Fanconi anaemia and Down syndrome (Table 1). The distribution of AML subtypes was similar in adults and children (Table 1). Patients with AML-MR were significantly older than patients with APL and the other subtypes of AML as a group (p = 0.026) (Figure 2a).

The median OS for all patients was 5 months (IQR 1–50). The 3-year OS was 19% (17/90).

For adult patients (≥ 18 years), the median OS was 3 months (IQR 1–18) with a 3-year OS of 15% (11/73) (Figure 3b). For adult patients (≥ 14 years, as per institutional division), the median OS was 4 months (IQR 1–18) with a 3-year OS of 14% (11/77). The median OS for the age group 14 years – 59 years was 7 months (IQR 1–22) with a 3-year OS of 21% (10/47), and the median OS for the older adults ≥ 60 years was 2 months (IQR 1–5) with a 3-year OS of 3% (1/30) (Figure 3a). Early mortality (death within 30 days of diagnosis) was 29% (22/77) for all adult patients (14 years – 82 years), 26% (12/47) in the younger adults (14 years – 60 years) and 33% (10/30) for older adults (≥ 60 years).

For children (< 18 years), the median OS was 19 months (IQR 4 – not reached) with a 3-year OS of 35% (6/17) (Figure 3b). In the paediatric patients ≤ 13 years, the median OS had not been achieved over the study period, and 46% (6/13) were alive at 3 years (Figure 3a). Early mortality was 23% (3/13).

Age had a significant impact on OS, with older patients experiencing lower OS than younger patients (Figure 3a and b). Survival was similar in males and females (p = 0.157) (Figure 3c). Survival was also better in patients with a white cell count (WCC) < 50 × 109/L (p < 0.001) (Figure 3d) and platelet counts > 80 × 109/L (p = 0.004) (Figure 3e).

The median OS of APL patients was 11 months (IQR 1–58, n = 19) and the 3-year OS was 37% (7/19) (Figure 3f). The 7-day mortality rate was 11% (2/19), and the 30-day mortality rate was 16% (3/19).

Almost half of the patients (47%, 9/19) had high-risk disease, six patients (32%) were intermediate risk and four patients (21%) were low-risk. The Sanz risk classification did not impact survival (p = 0.951) (Figure 3h).

The median OS of AML-MR patients was 2 months (IQR 1–7, n = 17). The 3-year survival was 6% (1/17). Patients ≥ 60 years old made up 59% (10/17) of those with AML-MR and would not have been eligible for intensive chemotherapy (Figure 2a).

Only 38 patients had sufficient genetic investigations (karyotype and/or NPM1 and FLT3 mutation testing) for ELN or BFM risk stratification (Table 1). The age of the patients with adverse risk was higher than those with favourable risk (p = 0.006) (Figure 2b). The adverse risk group had a significantly increased risk of death (p = 0.003, n = 57) (Figure 3f). In patients < 60 years, 36% (9/25) had adverse risk. There was no significant difference in survival between risk categories when only patients < 60 years were assessed (p = 0.206, n = 43), although similar trends in survival were observed (Figure 3g).

Increasing age (HR 1.022, 95% CI 1.008–1.037), higher WCC (HR 1.011, 1.005–1.017) and lower platelet count (HR 0.995, 95% CI 0.991–0.999) were independently associated with poor survival (Table 2). Secondary AML and risk stratification were not independent predictors of survival. Cox regression of only the 38 patients with risk stratification also did not reveal risk stratification as an independent predictor of survival, given the strong effect of age (Table 3).

This study has confirmed that the demographics, subtype frequencies and OS of AML were different from those seen in high-income countries.

The median age at diagnosis for adults of 54 years is older than other South African studies, typically in the third decade or fourth decade, but still younger than the 6th decade described internationally.^{4,5,6,7,10,11,12} The median age of 2 years for children is lower than reported elsewhere in Africa, where the median age is 7 years – 9 years.^34,35,36 The female predominance in the adult patients is similar to a Brazilian study,³⁷ but contrasts with the male predominance generally reported internationally.^4,5,6,7

Acute myeloid leukaemia diagnosis was consistent throughout the seasons, unlike reports from Spain and the US.^13,14 It should be noted that the Spanish (n = 26 475) and US (n = 21 570) studies were large; thus, a larger South African study is required to assess whether there is a seasonality of AML in South Africa.

The most common subtypes were APL (21%) and AML-MR (19%), which were also seen in another South African study.¹⁰

Survival rates were similar for males and females, which contrasts with the US, where mortality rates are higher for males.⁹

Survival in AML has improved over the years with increased understanding of AML biology and access to treatment. The median OS of adults (>18 years) at our institution was 3 months, which is shorter than the 8.9 months reported in a French population-based study of adult patients diagnosed between 1980 and 2004.⁸ In a US study of patients from 2001 to 2018, the 3-year relative survival for patients 20 years – 39 years old was 63.4% and for 40 years – 59 years old was 46.9%, superior to the 21% 3-year OS of adult patients (14 years – 60 years) in our study.⁹ Similarly, in our study, the median OS of adults 14 years – 60 years was 7 months, and in older adults (≥ 60 years) was 2 months, which poorer than a study from Brazil investigating AML patients from 2003 to 2009, where patients ≤ 60 years had a OS of 12.4 months and those > 60 years had an OS of 8.2 months.³⁸ A more recent Brazilian study of patients from 2013 to 2020 showed a median OS of 38.8 months and a 5-year OS of 39.4%.³⁷

Despite investigating patients from around the same time period, a study of patients in Romania receiving intensive treatment from 2015 to 2021 showed an OS of 8.7 months, highlighting the variability of AML outcomes across geographies.³⁹ The early mortality of our adult patients (14 years – 60 years) of 26% was also higher than the 17.3% in the Romanian study.³⁹ The median OS of our adult patients was better than the 2 months seen in adult AML patients in Uganda,⁴⁰ and the crude median OS of 1.75 months in a Johannesburg laboratory-based South African study.¹⁰ Importantly, patients at our facility who were > 60 years old and/or not fit for intensive chemotherapy did not have access to hypomethylation agents. These agents have been shown to improve OS from 2 months if untreated, to 6 months if treated, with the greatest benefit seen in those under 70 years achieving a median OS of 10 months.⁴¹

In paediatric patients (< 18 years), the median OS was 19 months, which was better than other African studies (2.6 months – 4 months), and similar to an Indian study of patients < 18 years (14.6 months).^34,42,43 Compared to European patients ≤ 14 years with a 5-year relative survival of 70.5%,⁴⁴ the 3-year OS of 46% for our paediatric patients ≤ 13 years highlights that there are several obstacles faced in a resource limited setting to achieve improved outcomes in paediatric AML. Inferior outcomes of paediatric AML in low-income and middle-income countries have been mainly attributed to abandonment of treatment, early death, treatment-related mortality and low salvage rates after relapse.^45,46

Further studies are required in the South African public sector in both adult and paediatric patients, to assess the reasons for the inferior OS. These may include high rates of infection-related mortality, limited access to intensive care and blood product support, barriers to care for patients living in rural areas, a lack of access to comprehensive diagnostic testing and MRD monitoring, as well as limited availability of novel therapeutic agents and stem cell transplantation for older patients in the public sector.

The OS of APL appears to vary significantly across South Africa. The proportion of high-risk APL between studies may account for some of the variation in outcomes between centres. While the median OS in our study was 11 months with a 3-year OS of 37% (47% high-risk, n = 19), a Johannesburg study showed a crude median OS of 1.5 months (51.2% high-risk, n = 43), and a study from Groote Schuur Hospital in Cape Town reported a 3-year OS of 76.5% (39% high-risk, n = 69), the latter having similar survival to that of high-income countries (28% high-risk, n = 1095).^10,47,48 Our study showed a 7-day mortality rate of 13% and a 30-day mortality rate of 19%, which was higher than the 7% and 13%, respectively, seen at the other Cape Town centre, but lower than the 18.5% and 33.3%, respectively, seen in a Bloemfontein study (56% high-risk, n = 27) and the 31.1% and 46.9%, respectively, seen in the Johannesburg study.^10,47,49 A larger study with a more in-depth analysis of APL at our centre is required to assess the high early death rate, including factors such as timely diagnosis and initiation of supportive care, delays in administering the first dose of ATRA because of transport issues from rural areas, and the availability of blood product support at peripheral hospitals to manage the associated coagulopathy.

In our study, age was an important independent risk factor for OS. The prognostic significance of risk stratification was not fully appreciated compared to the significant impact of age on survival. It is worth noting that the median age of the adverse risk group was older than that of the intermediate- and low-risk groups.

This highlights that older patients are not only more likely to have a generally poorer performance status and not receive intensive chemotherapy, but also have more adverse genetic abnormalities, which contribute to a worse prognosis.⁵⁰ That being said, addressing cost barriers and expanding access to effective therapies, such as venetoclax and hypomethylating agents may improve outcomes for this group of patients. The prognostic risk stratification would also likely have been adjusted if an next-generation sequencing (NGS) myeloid was available during the study period. In addition, high WCC was found to influence OS, which has been previously reported.^39,51,52

The prognostic significance of pretreatment platelet count is less robust, with some studies showing improved survival at low platelet counts, while others show better survival at higher platelet counts, as seen in our study.^53,54,55

The small number of adult AML patients with an available karyotype and/or molecular testing affected the subtype classification of AML and risk stratification, with only 57 patients and 38 patients with sufficient data, respectively. We did not have any patients that had AML with NPM1 mutation, but this may be attributed to suboptimal testing practices, rather than absence of this subtype at our facility. The incompletely investigated patients with CN-AML may be attributed to the karyotype results usually only being available after induction chemotherapy is complete. We recommend establishing an algorithmic approach for the work-up for an AML patient so that an inclusive panel of mutations can provide more comprehensive insights into the effect that AML genetics has on outcome in South Africa. Current practice at our institution has since evolved with the implementation of more comprehensive risk stratification of patients, facilitated by the availability of a myeloid NGS panel.

There were too few paediatric patients with AML over the study period for meaningful survival subanalysis.

Being a laboratory-based study, the reasons for inferior outcomes, including the high number of early deaths in adults, especially APL patients, could not be elucidated. A clinically focussed study is needed to further elucidate the contribution of clinical presentation and management aspects on outcomes.