This analysis included 206 de novo B-ALL pediatric Mexican patients from 10 different states of Mexico ( Supplementary Table 1A displays the number of patients by state). Patients were treated at 18 public health institutions ( Supplementary Table 1B ). The diagnosis was established between January 2018 and April 2023 by pediatric hematologists/oncologists according to clinical features: cell morphology, immunophenotype, and genetics, as defined by the 2008 WHO classification of lymphoid neoplasms (31). The patients had a range of 0 to 17 years, and only 3 of them were younger than one year. The risk classification was established as defined by the Children’s Oncology Group (COG) and the National Cancer Institute (NCI) (32). The presence of at least one of the following clinical features was considered as high-risk: age at diagnosis ≥10 years, WBC counts higher than 5x10⁴ count/µL, positive minimal residual disease (MRD) on day 28, presence of nervous system infiltration, treatment failure response, and identification of any high-risk gene fusion (when that information was available). Cytogenetic results were unavailable for most patients. Before treatment initiation, 3–5 mL of bone marrow was collected into EDTA tubes at their respective medical institutions and sent to the Laboratory of Genomic Diagnosis at INMEGEN. Clinical data were collected from medical charts, including the child’s sex, age, white blood cell count in the peripheral blood, and percentage of blasts in the bone marrow at diagnosis and treatment protocol. The median follow-up of these patients was 1.4 years.

DNA was isolated using the Maxwell® RSC Instrument (Promega Corporation, Madison, WI, USA). DNA concentration and quality were evaluated using a NanoDrop® 2000 spectrophotometer (Thermo Fisher Scientific Inc., Waltham, MA, USA) and a Qubit® 4 Qubit 1X dsDNA HS Assay Kit (Thermo Fisher Scientific Inc., Waltham, MA, USA).

A customized panel was designed to identify genetic alterations in CDKN2A, CDKN2B, PAX5, ERG, and IKZF1. Panel synthesis was performed using an Archer Dx (ArcherDX, Inc., Boulder, CO, USA). Library preparation was performed following the manufacturer’s instructions. Libraries were sequenced on the NextSeq 500/550 using the High Output Sequencing Reagent Kit v2.5 (300 cycles) (Illumina, Inc., San Diego, CA, USA). Bioinformatics analysis was performed using the Archer Suite Analysis v5.1.3 software (ArcherDX, Inc., Boulder, CO, USA) using the human reference genome GRCh37.p13/hg19. The bioinformatics pipeline allowed the identification of single nucleotide variations, small insertions/deletions, copy number alterations (CNAs), and structural variations (SV) (this last only for the IKZF1 gene). According to the Archer Dx bioinformatics pipeline (ArcherDX, Inc., Boulder, CO, USA), SV occurs when individual reads contain nucleotide sequences that are aligned to different genome regions. The length of the SV was determined by observing the breakpoint positions of the partners identified in the event and subtracting the differences in their genomic positions (User Manual Archer Analysis 6.0 CS001). Mutations were considered if the variant allele fraction (VAF) was ≥ 5%. A minimum 500x-fold depth coverage was required for all targeted regions in the panel.

NGS-predicted IKZF1, CDKN2A/2B, and PAX5 deletions were confirmed by MLPA using SALSA Probemix P335 (ALL-IKZF1) according to the manufacturer’s protocol (Holland, Amsterdam, Netherlands). This probe also allowed for the identification of PAR1 deletions, as described previously (3). The P327 iAMP21-ERG probe mix was used to confirm ERG deletions. Fragment separation was performed on a 3500 Genetic Analyzer (Thermo Fisher Scientific Inc., Waltham, MA, USA). The raw data files were imported and analyzed using Coffalyser.Net™ software (MRC Holland, Amsterdam, Netherlands).

The IKZF1^{plu s} profile was defined based on the Stanulla et al., 2018 report (3). Briefly, the patient was considered IKZF1^plus positive if IKZF1^DEL co-occurred with CDKN2A^DEL , CDKN2B^DEL (homozygous deletion only), PAX5^DEL , or pseudoautosomal region 1 (PAR1) deletion. As it has been reported that ERG^DEL mitigates the adverse prognosis associated with IKZF1 deletions (12), patients with ERG^DEL were excluded from the IKZF1^plus group. The NO-IKZF1^plus group included any other patient, regardless of whether they were IKZF1^MUT or IKZF1^WT .

Measures of central tendency were used to describe continuous variables related to the patient’s clinical and demographic features. The χ² test or Fisher’s exact test was used to assess the relationship between categorical variables. In contrast, the nonparametric Mann–Whitney U test was used to analyze the association between continuous variables, where p <0.05 was considered statistically significant. OS was estimated using the Kaplan-Meier method. The log-rank test was used to evaluate the differences between survival distributions with a 95% confidence interval (CI). The Cox regression model was used to perform a multivariate analysis, with adjustments made for other risk factors such as sex, age at diagnosis, WBC count, percentage of blast cells, and risk stratification at diagnosis. OS was calculated from the day of diagnosis until either the last follow-up or death from any cause. Patients who did not experience any event were censored at the last follow-up visit. Those who did not attend the follow-up were censored at the date of the last known contact. Statistical calculations were performed using the RStudio software version 4.3.1, and the data were visualized using the ggplot2 package version 3.4.3.

The IKZF1^plus profile was present in 21.8% (45/206) of the evaluated patients. Gene deletion combinations within the IKZF1^plus group are shown in Table 2 . In our series, three patients were excluded from the IKZF1^plus profile because of ERG^DEL .

The patients were stratified according to the presence of IKZF1^plus, and the clinical feature distribution was evaluated between the groups ( Table 3 ). Age at diagnosis showed significant differences between the groups, with patients with IKZF1^plus being older (p=0.04). Additionally, a trend toward higher risk classification was observed within the IKZF1^plus group. No significant differences were found in sex distribution, percentage of bone marrow blasts, or WBC count at diagnosis between the two groups.

The impact of IKZF1^plus on OS at five years was evaluated in 168 patients for whom information for OS calculation was available (35 IKZF1^plus vs. 133 NO-IKZF1^plus ). The IKZF1^plus group showed significantly reduced OS at 5 years compared to the NO-IKZF1^plus group (p=0.009, Figure 1A ). A significantly worse OS was also observed for the high-risk-IKZF1^plus -positive subgroup when patients were stratified by the IKZF1^plus status and the risk category (p=0.02, Figure 1B ).

We conducted a Cox regression multivariate analysis to examine the impact of the IKZF1^plus profile, taking into account additional clinical factors such as patient age, sex, white blood cell count, percentage of blast cells, and the risk classification at the time of diagnosis. The analysis revealed that having the IKZF1^plus profile significantly increases the risk of death by 3.7 times (HR = 3.7227, p = 0.02), highlighting its role as an independent poor outcome-predicting factor. The other variables included in the model did not demonstrate a statistically significant effect on the outcome.

The types of mutations identified in each gene and their frequencies are shown in Figure 2 . IKZF1 was altered (IKZF1^MUT ) in 35.9% (74/206) of patients studied ( Supplementary Table 3 for sequence mutations and Supplementary Table 4 for CNAs). Deletions were the most common alterations, representing 86.5% (64/74) of positive cases. Deletion affecting exons 4–7 (IK6 isoform) was the most common (39%, 25/64), followed by complete gene deletion (31.2%, 20/64). Moreover, a deletion spanning exons 1–8 was present in 7.9% (5/64) of cases ( Figure 3A ). SV was also detected in 54.7% of the patients with deletions. In contrast, only SVs were identified in seven individuals; interestingly, in six of them, the VAF was < 10%.

Sequence mutations were identified in 9 cases; in six of them, a concomitant deletion or SV was also present. Ten mutations were identified in nine cases: four missense mutations, four nonsense mutations, and two frameshift mutations, most of which were located in the DNA-binding domain coding region. In two patients, p.Asn159Tyr (N159Y) was identified without additional IKZF1 alterations.

At least one PAX5 alteration (PAX5^MUT ) was identified in 26.6% (55/206) of patients ( Supplementary Tables 3, 4 ). These include CNAs caused by complete or partial gene deletions, gains, and small sequence mutations. PAX5^DEL was the most common alteration, representing 75% (41/55) of PAX5^MUT . In most cases, 63.4% (26/41), several exons were deleted, most commonly from exon 2 to 5 or from exon 2 to 6 (14%, 6/41 of each). In two cases, only one exon was deleted (exon 1 or exon 7). The whole gene (exons 1-10) was deleted in 36.6% (15/41) of cases. PAX5^igAMP was identified in four patients; in two of them, additional genetic material was present in exons 2-5, one in exon 5, and one in the whole gene ( Figure 3B ; Supplementary Table 4 ).

A total of 17 PAX5 sequence mutations were identified in 15 patients: four nonsense mutations, five missense mutations [one p.Pro80Arg (P80R)], seven frameshifts, and one intronic variant ( Supplementary Table 3 ). In eight of them, a second alteration in PAX5 was also present: four deletions, two sequence mutations, and two gains of material in PAX5.

CDKN2A^DEL was identified in 35.92% (74/206) of the cases; in most cases (65/74), CDKN2B was entirely or partially deleted ( Figures 3C, D ). CDKN2A homozygous deletion was observed in 40 patients; in 30 of them, CDKN2B also showed homozygous deletion. The most frequent CDKN2A^DEL comprised exons 1 to 3 and was identified in 85.1% (63/74) of the patients, followed by the deletion of exon 1, which was identified in 14.9% (11/74) of CDKN2A^DEL patients. For CDKN2B, exon 1 and 2 deletions were identified in 87.7% (57/65), followed by exon 2 deletion in 12.3% (8/65) of the CDKN2B-positive cases ( Supplementary Table 4 ). In six cases, sequence variants were identified in CDKN2A (four cases) and CDKN2B (two cases) ( Supplementary Table 3 ).

ERG^MUT was identified in 8.7% (18/206) of cases ( Supplementary Tables 3, 4 ). Among the ERG^DEL patients (13/18), involved various exons. These deletions affect the region expanding exons 2 to 11, encoding the ERG protein’s pointed domain (PNT) or alternative domain (AD) of the ERG protein. Sequencing variants were identified in five cases, three of which were located in the erythroblast transformation-specific domain (Ets) of the protein ( Figure 3E , Supplementary Table 3 ).

Deletion of the PAR1 region was assessed by MLPA in patients with IKZF1^DEL . We identified only ten patients with PAR1 deletions ( Supplementary Table 4 ).

Sex distribution, age at diagnosis, WBC count at diagnosis, percentage of blasts, and risk assignment were compared between patients positive and negative for IKZF1, PAX5, CDKN2A/2B, and ERG mutations ( Supplementary Tables 5A–D ). The frequency of high-risk patients and of those older than 10 years at diagnosis was significantly higher in the IKZF1^MUT group (p=0.02 and p=0.002, respectively). A higher proportion (37.7%) of patients with CDKN2A/2B^MUT had a WBC count in the range of 20,000–100,000 cells than CDKN2A/2B^NEG patients (22.4%) (p=0.01). Additionally, a significantly higher proportion of patients had > 50% blasts at diagnosis in the CDKN2A/2B^MUT (98.6%) group than in the CDKN2A/2B^NEG group (89.7%) (p=0.05) ( Supplementary Table 5B ). In the case of PAX5^MUT , a higher percentage of cases was associated with high-risk classification at diagnosis, but the difference was not statistically significant (p=0.06).

A significant decrease in OS was observed in the IKZF1^MUT (p=0.02) ( Supplementary Figure 1A ) and CDKN2A/2B^MUT (p=0.006) ( Supplementary Figure 1B ) groups. Significant differences (p=0.02) in OS were identified between the homozygous CDKN2A^MUT and heterozygous CDKN2A^MUT vs. CDKN2A^NEG patients. ( Supplementary Figure 1C ). No significant effect on OS was observed for PAX5^MUT or ERG^MUT ( Supplementary Figures 1D, E ).

To determine whether the observed negative impact of CDKN2A/2B^MUT and IKZF1^MUT mutations on OS was conferred by IKZF1^plus -positive patients in both groups, we excluded IKZF1^plus patients from the OS analysis. After eliminating these patients, no significant differences in OS were observed between the CDKN2A/2B^MUT or IKZF1^MUT groups and those without mutations in these genes ( Supplementary Figures 2A, B ).

The frequency of IKZF1^MUT in this cohort (35.9%) was higher than that reported for Mexican patients. Ayón-Perez et al. found that 20.6% (36 cases) (37), while Rosales-Rodríguez et al. (38) detected 27% (63 cases) both using MLPA. The frequency of IKZF1^MUT in our study was still higher (31.1%) than previously reported in Mexicans if only the deletions were considered in the IKZF1^MUT count. IKZF1^MUT frequencies in other populations were also lower: American (28.6%) (9), Brazilian (19.3%) (39), and Argentinian (9.2%) (5). The differences in the sensitivity of the methodology contributed to the increased frequency detected in our study. In most previous reports, MLPA was used; therefore, SV present in a low proportion could not be observed. In fact, seven patients in our study had SV with a VAF < 10%; concomitant deletion was not detected in these patients. Therefore, they would have been considered “negative” if MLPA had been used as the detection method.

In most patients with IKZF1^DEL , deletion of exons 4–7, which encode the dominant negative IK6 isoform, was identified. Whole gene deletion, resulting in the loss of expression of the wild-type allele, was the second most frequent deletion. Four patients showed deletions in the noncoding exon 1. Deletions involving the 5’ region of IKZF1 also result in haploinsufficiency owing to a significant reduction in IKZF1 mRNA expression (40). IKZF1 is intolerant to variants that cause loss of function, according to the constraint metrics obtained from the gnomAD exome database (41).

The IKZF1 ^N159Y mutation was present in less than 1% of our cases (two cases, 0.9%), and a similar low frequency was found in an analysis of 1988 pediatric B-ALL cases (0.4%) (42). No other IKZF1 alterations were present in these two patients, suggesting retention of the non-mutated IKZF1 allele, as previously reported in this subgroup of patients (42). This mutation is located inside the DNA-binding domain and induces nuclear mislocalization and intercellular adhesion. The transcriptomic profile of the IKZF1 ^N159Y positives was different from that observed in other patients with alterations in IKZF1, suggesting that IKZF1 ^N159Y could define a new subtype of B-ALL (42). The upregulation of genes involved in oncogenesis, chromatin remodeling, and signaling has been identified in patients with IKZF1 ^N159Y (43).

IKZF1^DEL was associated with older age at diagnosis, which is in accordance with the results of previous studies (9, 44). IKZF1^DEL was associated with poor outcomes in several studies (9–11). In our patients, a significant decrease in OS was also observed for IKZF1^MUT (p=0.02). Still, it seems to have been mainly contributed by the IKZF1^plus patients present in the IKZF1^MUT group since significance was lost after removing the IKZF1^plus patients. In previous reports, the contribution of IKZF1^plus to the poor outcome observed with IKZF1^MUT was not independently evaluated.

PAX5^MUT was less frequent in Mexican patients than in other ethnic groups. PAX5^MUT has been found in approximately one-third of non-Hispanic B-ALL pediatric cases, according to studies performed in the USA and Netherlands (9, 45). In this study, the proportion of PAX5^MUT was only 26.6%, considering all alterations (deletions, gain, and small sequence mutations). The frequency of PAX5 fusion-positive patients increased to only 27.7%, which is still lower than that reported in other studies (46). A previous study in Mexican patients found that 15.9% PAX5^DEL by MLPA (38) was lower than that reported in Brazilians (25.2%) using the same methodology (39).

Monoallelic deletions affecting the paired domain were the most frequently identified PAX5 alterations, similar to the findings in other groups (9, 45). They involve small portions of the gene or cause whole-gene loss, impairing the DNA-binding capacity of the PAX5 protein. Mouse models have shown that haploinsufficiency of Pax5, caused by monoallelic deletion, confers susceptibility to B-cell transformation. However, other oncogenic events that act synergistically are required for leukemic transformation (47). In our cohort, PAX5^DEL was concomitant with the BCR::ABL1, Ph-like-related fusions, and EBF1 mutations. The activation of STAT5 and mutations in EBF1, JAK3, and BCR::ABL1 have also been identified as cooperative events in PAX5^DEL mice (48, 49).

Extra copies of PAX5 regions were identified in four cases, representing 1.9% of the total and 7.4% of the PAX5-positive group. An extra chromosome 9 would be present in one of them, as the other genes in chromosome 9 included in the NGS panel also showed more than two copies. No cytogenetic data were available for this patient to confirm the presence of extra Chr 9. PAX5^igAMP was the most likely alteration in the three other patients, with a gain in genetic material observed for the exons encoding the DNA-binding and octapeptide domains, as reported previously (50).

PAX5^igAMP has been associated with male sex, age at diagnosis older than ten years, and high-risk stratification. This alteration is frequently associated with CDKN2A/2B^DEL (82%), and appears to be mutually exclusive with other significant risk-stratifying genetic lesions (50). Despite the small number of patients with PAX5^igAMP in our analysis (3 cases), the clinical features were similar to those reported: three were males, 2/3 were older than ten years, and 1/3 had nine years; all were classified as high-risk at diagnosis. CDKN2A/2B^DEL was present in all cases, and BCR::ABL1 was the only fusion identified (one case). Although the functional consequences of amplification have not yet been established, it has been suggested that increased copies of the DNA-binding region may alter binding to PAX5 target genes, leading to dysregulated B-cell differentiation and transformation (51). PAX5^igAMP was associated with a high incidence of relapse (50). We did not evaluate the impact of PAX5^igAMP on OS because of the small number of patients.

It has been suggested that some sequencing mutations may serve as initiating rather than secondary cooperative events in leukemogenesis (42). PAX5 point mutations have been identified in 7–10% of pediatric cases of B-ALL, similar to the frequency in our patients (7.3%), being the second most common PAX5 type of alteration. Fourteen mutations were identified, mainly located in the paired or transactivating domains. The most common consequence was the loss of function due to a frameshift or premature stop codon (7/14). Only seven of them have been previously reported in the COSMIC database (as of October 2023), and two previously described cases were recurrent in our patients: p.Ala322ArgfsTer19 (three cases) and p.Trp358Ter (two cases), both of which affect the transactivation domain. The frequency of PAX5^P80R in our cases was lower than that previously reported. Only one case (0.5%) was identified, in contrast to the 2–6% observed in other studies (42, 52–54). In accordance with previous observations, this case was negative for fusions or iAMP21 and had an additional PAX5 loss-of-function mutation affecting the other allele. Martínez-Anaya et al. reported PAX5^P80R in 1.4% of the patients; however, only Ph-like Mexican pediatric patients were analyzed (55). In seven cases with sequence mutations, both PAX5 copies were affected, with a deletion, gain, or a second point mutation in the other allele.

PAX5 alterations can function as cooperative events or germline-initiating genetic lesions in B-ALL patients (23). In most of our cases, the VAF of PAX5 sequence mutations was lower than expected, considering the percentage of blast cells present in the bone marrow sample analyzed, suggesting that PAX5 sequence mutations are secondary events that occur only in a subpopulation of malignant cells. Additionally, none of the sequencing mutations identified in our study has been reported as germline variants according to the HGMD professional database (as of October 2023) or had a VAF suggesting a germline origin.

As observed in another series of B-ALL patients, CDKN2A/2B^DEL was the most frequently identified CNA (35.9%) in Mexican patients. Homozygous deletion of both genes occurred in a high proportion of the cases (40.5%). As previously reported, none of the patients had CDKN2B^DEL without CDKN2A^DEL (3). The frequency of CDKN2A/2B^DEL was higher than that reported by Rosales-Rodríguez et al. (31.7%) in Mexican patients (38). Lower frequencies have also been reported in other Latin American and Asian populations, including Colombia (25%) (56), Brazil (31.8%) (39), China (20.2%) (15), and India (19.8%) (57). In our study, CDKN2A/2B^DEL was associated with higher blast counts at diagnosis and high-risk features. Most patients with this alteration were classified as high-risk, although the difference was not statistically significant.

Sequence variants of CDKN2A/2B, including CDKN2A (1.9%) and CDKN2B (0.97%), occur in a very low proportion of patients with B-ALL. Three CDKN2A sequence variants, p.Ile49Thr, p.Val51ProfsTer89, and c.151-6_151delinsC, had a VAF close to 50% and were classified as likely pathogenic according to ACMG criteria (58). However, p.Ile49Thr is probably of germline origin since it has been previously identified in the Latino population database (f=0.00451, 1456/34586, gnomAD exomes database, as of October 2023). Although it is listed in ClinVar (ID 127523) with conflicting classifications (pathogenic vs. uncertain significance), functional studies support its role in pathogenicity (59). However, this effect appears more moderate than other known pathogenic variants of CDKN2A (60). The variant has been reported to be pathogenic in families with melanoma and pancreatic cancer (60). It has been identified several times in Mexican women with breast cancer who fulfill genetic risk criteria. Ten patients with cancer carrying this variant, identified in our laboratory, reported no family history of leukemia or other hematological malignancies during routine evaluation for cancer predisposition syndromes. No history of cancer was present in the family of the B-ALL patient carriers of this variant. The variant p.Val51ProfsTer89 produces a loss of function due to a frameshift, and c.151-6_151delinsC is a deletion insertion that affects the canonical splicing site in exon two and may cause nonsense-mediated decay. These variants are not listed in the HGMD or ClinVar databases (as of October 2023), are absent from population databases, and have not been reported as germline variants in cancer syndromes. None of the patients had a family history of cancer. As DNA obtained from non-hematopoietic tissue was unavailable, variants’ germline or somatic origins could not be clarified.

CDKN2B p.Asn44Ser was identified in two cases, with a VAF of approximately 50%. It is classified as having uncertain significance according to ACMG and has not been reported as a somatic mutation. It has a low frequency in Latino individuals. (f=0.0121%, 4/33098, gnomAD exomes, as of August 2023), and was absent from the Mexican database incorporated into the Franklin By Genoox platform (470 individuals). The allele frequency in our cohort of B-ALL patients was significantly higher than in gnomAD patients (0.4% in B-ALL vs. 0.0121% in gnomAD exomes, X² = 22, p=2.7E-06). As both patients were from Oaxaca State, we can speculate that this variant could be frequent in this region of Mexico. However, a possible association with an increased risk of leukemia development cannot be ruled out without further investigation.

OS was significantly decreased (p=0.006) in the CDKN2A/2B^MUT group, but similar to what happened for the decreased OS in the IKZF1^DEL group, the significance was lost after removing the IKZF1^plus patients. CDKN2A/2B^DEL has been correlated with inferior 3-year event-free survival and 3-year OS rates in B-ALL patients (15). Other researchers have concluded that CDKN2A/2B^DEL is associated with an increased probability of relapse and death (17, 57, 61, 62), but the contribution of IKZF1^plus patients to these outcomes has not been evaluated.

No conclusive data exist regarding the contribution of CDKN2A/2A^DEL heterozygous or homozygous status. In our study, CDKN2A/2B^DEL heterozygous patients showed a significant decrease in OS (p=0.02) compared with homozygous and wild-type patients. However, Feng et al. (15) showed that patients with biallelic deletions had a worse 3-year EFS than those without. However, other authors have reported that homozygous CDKN2A/2B^DEL is not associated with poor prognosis in childhood B-ALL (63).

In our cohort, somatic mutations in ERG were identified in 8.7% of patients, with most presenting deletions (72.2% of the ERG^MUT cases) that affect one or more exons encoding the ERG protein’s pointed domain of the ERG protein or the alternative domain. All sequence variants identified in ERG are probably somatic, considering the VAF, and are located mainly in the erythroblast transformation-specific domain. According to the COSMIC database (as of October 2023), these domains are reported to be mutated in hematopoietic neoplasms.

Patients with ERG^MUT had better OS, although statistical significance was not reached because of the small sample size. None of the patients with ERG^DEL experienced mortality, supporting the reported association between ERG^DEL and favorable outcomes (12, 13).

The present study has several limitations that must be considered when interpreting the results. It was a multicenter retrospective analysis of a heterogeneous group of patients. Treatment protocol and hospital-related factors would have influenced OS. The patients were treated according to six protocols, the only target therapy available was imatinib, and it was used only in the BCR::ABL1 positive patients. Due to the lack of molecular testing, Ph-like patients were not identified and did not receive target therapy, which would affect their OS. Economic resources and hospital infrastructure may vary among the 18 hospitals where the patients were treated. Cytogenetic/FISH characterization and MRD determination are not routinely performed in all Mexican public institutions, and this information was unavailable for these patients. Despite being one of the most extensive series evaluating molecular alterations in pediatric B-ALL, the number of patients is still small, and clinical information was not available for 100% of the patients; thus, it reduces the statistical power to detect additional differences in the distribution of clinical features between mutation-positive and mutation-negative patients or between subgroups of IKZF1^plus patients or to evaluate the impact of specific mutations on OS. Finally, additional biological and socioeconomic factors that were not assessed in the present study could have contributed to the poor patient outcomes.