Evaluation of a Fever-Management Algorithm in a Pediatric Cancer Center in a Low-Resource Setting

Abstract

Background:

In low- and middle-income countries (LMICs), inconsistent or delayed management of fever contributes to poor outcomes among pediatric patients with cancer. We hypothesized that standardizing practice with a clinical algorithm adapted to local resources would improve outcomes. Therefore, we developed a resource-specific algorithm for fever management in Davao City, Philippines. The primary objective of this study was to evaluate adherence to the algorithm.

Procedure:

This was a prospective cohort study of algorithm adherence to assess the types of deviation, reasons for deviation, and pathogens isolated. All pediatric oncology patients who were admitted with fever (defined as an axillary temperature > 37.7°C on one occasion or ≥ 37.4°C on two occasions 1 h apart) or who developed fever within 48 h of admission were included. Univariate and multiple linear regression analyses were used to determine the relation between clinical predictors and length of hospitalization.

Results:

During the study, 93 patients had 141 qualifying febrile episodes. Even though the algorithm was designed locally, deviations occurred in 70 (50%) of 141 febrile episodes on day 0, reflecting implementation barriers at the patient, provider, and institutional levels. There were 259 deviations during the first 7 days of admission in 92 (65%) of 141 patient episodes. Failure to identify high-risk patients, missed antimicrobial doses, and pathogen isolation were associated with prolonged hospitalization.

Conclusions:

Monitoring algorithm adherence helps in assessing the quality of pediatric oncology care in LMICs and identifying opportunities for improvement. Measures that decrease high-frequency/high-impact algorithm deviations may shorten hospitalizations and improve healthcare use in LMICs.

Introduction

Infectious complications are major contributors to mortality among pediatric patients treated for cancer in low- and middle-income countries (LMICs).^1,2 Specifically, inconsistent or delayed management of infections is a driver of poor outcomes.^1,3–6 In high-income countries (HICs), guidelines and algorithms have reduced management inconsistencies and expedited appropriate care.^7–9 Fever-management algorithms for oncology patients are commonly used in these settings. Although the limitations on staffing and specialist support in LMICs might suggest that such algorithms would be particularly useful in those countries, the differences in health delivery systems and resource constraints preclude the use of identical guidelines in all LMICs. Nevertheless, practical recommendations are required for these countries, where an estimated 80% of all children with cancer live.¹⁰ Accordingly, the International Society for Pediatric Oncology (SIOP) recommends that LMIC centers treating patients with cancer develop local fever-management protocols, although it does not endorse specific therapeutic or design approaches.¹ The Sociedad Latinoamericana de Infectología Pediátrica (SLIPE) has published a consensus guideline prepared in collaboration with healthcare workers in LMICs, but this guideline is still based on experience in HICs.³ The primary objective of our study was to measure healthcare provider adherence to a resource-specific algorithm for managing febrile pediatric oncology patients in Davao City in the Philippines, which is defined by the World Bank as a lower middle–income country.

Methods

Study setting

The Southern Philippines Medical Center (SPMC) is a tertiary referral institution serving Davao City and southern portions of Mindanao Island. The SPMC pediatric cancer center (PCC) consists of a 25-bed inpatient ward, an outpatient clinic, and on-campus housing. Each year, it treats approximately 200 pediatric patients newly diagnosed with cancer. The PCC has ongoing collaborations with St. Jude Children’s Research Hospital (St. Jude) in clinical and quality-improvement activities. The PCC staff includes two pediatric oncologists, two hematologists, and 21 dedicated oncology nurses. Much of the direct patient care is provided by three or four pediatric residents, who complete 2 months on the oncology service between their second and third years of residency. At the beginning of each rotation, residents receive ward orientation with training in specific cancer-care protocols. Oncology providers make daily rounds, and overnight subspecialist support is available by telephone. The ALL treatment protocol uses a Children’s Oncology Group (COG) backbone adapted to a limited-resource setting. Chemotherapy drugs such as prednisone, vincristine, methotrexate, doxorubicin, L-asparaginase, mercaptopurine, and cyclophosphamide are given over a period of 2.5 years. During the course of treatment, patients are instructed to proceed directly to the PCC should they develop fever or symptoms suggestive of infection; however, some patients are admitted from the emergency department or outpatient clinic.

Algorithm development

An algorithm to manage fever in pediatric oncology patients was developed and piloted over a 13-month period starting in February 2013. The algorithm was created using a three-step process: (1) information gathering; (2) algorithm and consensus building; and (3) algorithm validation and pilot implementation. A combined SPMC/ St. Jude research team first conducted semi-structured interviews with medical personnel, reviewed medical records, observed care practices, and verified the pricing and availability of ancillary testing and antimicrobials. This investigation revealed that attending physicians, residents, and nursing team members had different definitions of fever, different perceptions of its urgency, and different approaches to managing it. For example, blood cultures were not routinely obtained, and antibiotics were not selected in a uniform manner because of supply problems, cost concerns, and knowledge limitations. To address these discrepancies, our team generated an algorithm by adapting published guidelines to local resources.^3–6 HIC clinical criteria were modified to devise a risk-stratification rubric to identify patients at high risk for serious infection on admission.^11–15 Management decisions were then dichotomized based on the patient’s risk profile. Patient risk and treatment response were recategorized every 2 days until day 7 of the febrile episode to direct further management. The algorithm was approved by PCC oncology and infectious diseases personnel in March 2013.

In the 1-year pilot implementation phase, all physicians and nurses providing inpatient care were trained in algorithm use by the infection preventionist during ward orientation. The algorithm was posted in the nurses’ station and incorporated into ward reference material, and the infection prevention team reviewed its use during weekly clinical meetings. Obstacles to algorithm implementation and changes in antimicrobial availability were discussed with the St. Jude team throughout the year, and the algorithm was continually amended based on feedback received. The revised algorithm was approved by the local team in March 2014 (Supplementary Figure S1).

Training in the revised algorithm

In March 2014, team members conducted mandatory on-site training sessions on the updated algorithm and blood-culture collection practices for all current healthcare providers of the PCC. Lectures were recorded to standardize the training for each group of rotating residents and for retraining purposes. The algorithm was posted in the nurses’ station, treatment rooms, outpatient clinic, and emergency department.

Monitoring adherence

A prospective study of algorithm adherence began in September 2014. All patients aged 18 years or younger with cancer diagnoses who received chemotherapy within 3 months of admission and were admitted with fever (defined as an axillary temperature > 37.7°C measured on one occasion or an axillary temperature ≥ 37.4°C measured on two occasions 1 h apart) qualified for inclusion. Patients who were afebrile on presentation but developed fever within 48 h of admission were also included.⁷ The protocol was approved by the Ethics Committee of the SPMC and by the St. Jude Institutional Review Board. The requirement for informed consent was waived, as the intervention fell within the scope of ongoing quality improvement.

Outcomes measured

Sociodemographic information, clinical and laboratory details of the presenting illness, and oncologic history were obtained for each patient at enrollment. The primary outcome measured was algorithm adherence, defined as the completion of any mandatory step of the algorithm. Secondary objectives included describing the types of deviation and reasons for deviation (nonadherence), along with the differences in the length of hospital stay, incidence of admission to the intensive care unit (ICU), and mortality between patients whose fever was managed according to the algorithm and those whose fever management did not follow the algorithm. The study focused on the first 7 days of hospitalization after fever onset, because the clinical course is too variable thereafter to be amenable to algorithmic management. Patients were identified for inclusion during morning rounds, and algorithm adherence was monitored prospectively by a nurse on the infection prevention team who was involved in developing and implementing the algorithm. When a deviation was noted, this individual approached practitioners to ascertain the reasons why the deviation occurred.

Statistical analysis

The study outcomes, along with demographic and clinical features of the study population, were summarized by descriptive statistics (i.e., medians and ranges for continuous data and frequency counts and proportions for categorical data). Chi-squared or Fisher’s exact tests (for sparse data) were used to assess two-way associations between categorical variables. The Cochran-Armitage test for trend was used to assess the association between the number of admissions (defined as 1, 2, or >2) and the occurrence of a deviation from the algorithm on any day. The Wilcoxon rank-sum test (for two independent groups) and Kruskal-Wallis test (for three or more independent groups) were used to compare continuous variables. Multiple linear regression was used to assess the joint effects of sociodemographic factors (age, gender, and income level), clinical factors (absolute neutrophil count, body mass index, symptoms at admission, pathogen isolation, and time between onset of fever and admission), types of deviation (stratification error, missed dose, missed culture, or failure to start vancomycin despite indications), and reasons for deviation (financial constraints, provider knowledge deficiency, identification of an infectious focus, or the drug being out of stock) on the duration of hospitalization. Duration of hospitalization was selected a priori as a surrogate marker for complications related to infection, as a review of the baseline data suggested that other objective outcomes, such as ICU admission and mortality, would be rare. The deviation type and reason for deviation are highly correlated; thus, two models were considered: one including types of deviation and covariates and another including reasons for deviation and covariates. The stepwise model selection method was used to select model variables with a significance level of 0.15 for inclusion. Analyses were conducted under the assumption that each admission was independent, when appropriate. A two-sided significance level of P < 0.05 was used for all statistical tests unless otherwise indicated. All statistical analyses were conducted using SAS Version 9.4 (SAS Institute, Cary, NC).

Results

Patients

Between September 1, 2014, and September 30, 2015, 141 febrile episodes were accrued representing 93 individual patients; Table 1 presents their sociodemographic and clinical characteristics. Of the patient episodes, 89% qualified for national insurance subsidization of up to 50% of hospital expenses based on the patients’ annual income. We found no association between the delay in hospital presentation (N = 117 episodes; median delay (IQR), 7 h (4.8–11.4); range, 0.2–368.9) and patient income level, household size, family educational level, or health insurance status. The median number of admissions for fever was 1 (range, 1–8), and 73 (78%) of the patients were admitted only once.

TABLE 1.

Characteristics of febrile episodes (N = 141)

Characteristics
Age (years), median (range)	5.5 (0.6 to 19.4)
Sex
Male	58 (41)
Female	83 (59)
Diagnosis/treatment phase
ALL/all phases	72 (51)
ALL/induction	20 (28)
ALL/consolidation	40 (56)
ALL/maintenance	12 (17)
AML/all phases	8 (6)
AML/induction	2 (25)
AML/consolidation	5 (63)
AML/off treatment	1 (13)
Rhabdomyosarcoma	16 (11)
Osteosarcoma	9 (6)
Retinoblastoma	8 (6)
Other	28 (20)
Highest level of education
Grade school	9 (6)
High school	82 (58)
College	48 (34)
Unknown	2 (1)
Health insurance
No	69 (49)
Yes	72 (51)
Area of residence
Northern Mindanao	16 (11)
Davao region	75 (53)
SOCCKSARGEN	39 (28)
Caraga	6 (4)
ARMM	5 (4)
Annual per capita income (pesos)a
<19,667	89 (63)
19,667 to 27,954	27 (19)
>27,954 to 35,941	10 (7)
>35,941 to 43,927	6 (4)
>43,927	5 (4)
Unknown	4 (3)
median (range), n = 137	14,400 (0 to 135,000)
No. of household members
≤5	80 (57)
>5	60 (43)
Unknown	1 (1)
median (range), n = 140	5 (3 to 13)
Height (cm), median (range)	106 (62 to 170)
Weight (kg), median (range)	16 (6 to 87)
BMI (kg/m2) for ages 2–20 years (n = 136)
Underweight (<5th percentile)	52 (38)
Normal (5th to <85th percentile)	73 (54)
Overweight (85th to <95th percentile)	9 (7)
Obese (≥95th percentile)	2 (1)
median (range)	14.7 (10.2 to 30.1)
Initial ANC (in cells/mm3)
<500	57 (40)
≥500	84 (60)
median (range)	821.4 (0 to 34,000)

Data are presented as n (%) unless otherwise indicated.

Abbreviations: ALL, acute lymphoblastic leukemia; AML, acute myeloid leukemia; ARMM, Autonomous Region in Muslim Mindanao; BMI, body mass index; ANC, absolute neutrophil count.

SOCCKSARGEN includes South Cotabato, Cotabato, Sultan Kudarat, Sarangani, and General Santos.

^a(monthly family income / number of household members) × 12

Algorithm adherence and deviation

Supplementary Figure S1 shows the algorithm for adherence evaluation. One or more deviations from the algorithm occurred in 50% of all febrile episodes (70 of 141) on day 0 (the first day of fever). A total of 259 deviations were charted during the first 7 days of admission in 65% of patient episodes (92 of 141). The most common deviation on day 0 was failure to perform blood cultures before antibiotic initiation, which occurred in 23% of episodes (33 of 141). No order was entered for 62% of missed cultures (21 of 34), and although the culture bottle was provided, the associated laboratory fee was unaffordable in 21% of cases (7 of 34) in which blood culture was missed (Supplementary Table S1). The number of patients with algorithm nonadherence dropped markedly on days 1 and 2, with a slight increase on day 3. Because fewer patients remained hospitalized beyond day 4, the subsequent decrease in episodes with nonadherence reflects the diminishing number of patients at risk (Supplementary Table S2).

Figure 1A shows the proportion of daily deviations of each type. As certain orders, such as those for radiology and laboratory tests, are routine only on day 0, the relative proportions of deviation represented by other order types increase on subsequent days. Failure to appropriately treat high-risk patients constituted 13% of deviations on day 0 and 29% of deviations on day 1. Missed doses of ordered antimicrobials accounted for 21% to 50% of deviations on days 1 through 6 of hospitalization. Figure 1B shows the reasons provided for nonadherence.

FIGURE 1.

Types of deviation observed (A) and reasons provided for deviation (B) according to day of hospitalization. (A) Deviation types are presented as a percentage of the total daily deviations. The trend line indicates the total number of deviations per day. Deviation types: Early discharge, discharge on a day other than indicated by the algorithm; Non-algorithm antibiotics, antibiotics other than those stipulated in the algorithm were prescribed; Broad spectrum, a patient categorized as low risk received antibiotics indicated for high-risk patients; Blood culture, blood culture was not performed; Missed dose, at least one dose of an ordered antimicrobial agent was not administered; Other lab, an indicated laboratory test other than blood culture was not performed; Stratification, a patient with high-risk characteristics was treated as a low-risk patient; Vancomycin, a clinician judged that a patient met criteria for administration of vancomycin, but this was not administered. (B) Reasons for deviation are provided as a percentage of the total reasons for deviation provided on the given day. The trend line indicates the number of reasons for deviation seen each day. Reasons for deviation: Nosocomial, provider treating for nosocomial infection; Knowledge, provider knowledge deficiency; Financial, inability to afford medicines or laboratory tests; Infectious source, provider treating for suspected infectious source.

The frequencies of presenting symptoms for all episodes are listed in Supplementary Table S3, and selected clinical outcomes are presented in Table 2. Patients defervesced within 48 h of admission in 48% of episodes with documented fever after admission (67 of 139), and ICU admission and mortality were infrequent. Organisms isolated from bacteremias or from alternate sites are listed in Table 3. Bacteria were isolated from 12 blood cultures. Two isolates testing positive for Staphylococcus hominis (17% of positive blood cultures) were considered to represent contaminants, and a third—a Bacillus species in a patient with abdominal symptoms—was of uncertain pathogenicity. Escherichia coli was isolated from three blood cultures and from a single urine specimen, with two of the three blood isolates and the single urine isolate being extended-spectrum beta-lactamase (ESBL)-producing variants.

TABLE 2.

Clinical outcomes

Outcome	N = 141 febrile episodes
Median duration of hospitalization in days (range)	8 (2–39)
Presentation of fever before admission	117 (83)
Median time in hours between fever and admission (range)	7.0 (0.2–368.9)
Fever after admission	139 (99)
Median duration of fever after admission in hours (range)	72 (24–168)
Mortality	5 (4)
Intensive care unit admission	7 (5)
Episodes with positive blood cultures	12 (9)
Infectious diagnoses
Pneumonia	63 (45)
Mucositis	20 (14)
Other	14 (10)
Ceftriaxone monotherapy changed to a different broad-spectrum agenta	33 (23)

Data are presented as n (%) unless otherwise indicated.

^aDefined as meropenem, cefepime, or piperacillin/tazobactam.

TABLE 3.

Frequency of pathogen isolation

Pathogen	Site	Frequency
Acinetobacter jeunii	Blood	1
Bacillus species	Blood	1
Chryseobacterium indologenes	Blood	1
Escherichia coli	Blood	3
Klebsiella pneumoniae	Blood	1
Pseudomonas aeruginosa	Blood	2
Staphylococcus hominis	Blood	2
Streptococcus pneumoniae	Blood	1
Escherichia coli	Urine	1
Ascaris lumbricoides	Stool	1
Entamoeba histolytica	Stool	5

Pneumonia was commonly reported but was not associated with poor outcomes. The presence of mucositis grade III on admission (P = 0.035), isolation of a pathogen (P = 0.005), diagnosis of a nonrespiratory infectious focus (P < 0.001), and use of a broad-spectrum antibiotic on any day (P < 0.0001) were all associated with significantly prolonged hospitalization on univariate analysis (Supplementary Table S4). Episodes with deviation on day 1 resulted in longer hospitalization than did episodes with no deviation on day 1 (10 versus 8 days; P = 0.004). Among those episodes with deviation on day 0, risk-stratification errors were associated with hospitalization that was significantly longer than that associated with other deviations (12 versus 7.5 days; P = 0.01). The Cochran-Armitage test revealed a significant increase in the proportion of patients with deviation on any day as the number of admissions per patient increased (P = 0.003).

In the multivariate model looking at the effect of deviation types and covariates on length of hospitalization, a stratification error [β (SE) = 6.03 (1.95); P = 0.003], a missed antibiotic dose [β (SE) = 4.05 (1.53); P = 0.009], pathogen isolation [β (SE) = 5.15 (1.77); P = 0.004], and male gender [β (SE) = 3.33 (1.19); P = 0.006] were the independent factors associated with prolonged stay. In the multivariate model looking at the effect of reasons for deviation and covariates, inability to afford indicated testing or medications [β (SE) = 3.77 (1.46); P = 0.011], pathogen isolation [β (SE) = 4.68 (1.84); P = 0.012], and male gender [β (SE) = 2.87 (1.21); P = 0.020) were the only independent predictors of prolonged stay (Table 4).

TABLE 4.

Multivariate regression model results for investigation of predictors of duration of hospitalization

	Parameter estimate	Standard error	P
Deviation-type model (N = 132, R2 = 0.19 )
Intercept	6.28	0.88	<0.0001
Sex (male vs. female)	3.33	1.19	0.006
Pathogen isolation (yes vs. no)	5.15	1.77	0.004
Ever a stratification error (yes vs. no)	6.03	1.95	0.003
Ever a missed dose (yes vs. no)	4.05	1.53	0.009
Deviation-reason model (N = 132, R2 = 0.12)
Intercept	7.03	0.90	<0.0001
Sex (male vs. female)	2.87	1.21	0.020
Pathogen isolation (yes vs. no)	4.68	1.84	0.012
Ever a financial reason (yes vs. no)	3.77	1.46	0.011

Discussion

Our experience reveals that even locally created algorithms face barriers to implementation. Some deviations from the algorithm, e.g., risk-stratification errors or missed antibiotic doses, were associated with prolonged hospitalization, as was inability to afford an indicated service. Identifying and averting key deviations may improve hospital resource utilization and the quality of care delivered in PCCs in LMICs, signaling the potential of fever algorithms as quality-improvement tools.

When we excluded predefined minor deviations (i.e., nonindicated use of broad-spectrum antibiotics and failure to perform urinalysis) from our analysis, febrile episodes with major deviations increased overall mortality, but this was not statistically significant (6% versus 0%; P = 0.075). Our findings are consistent with reports that clinical practice guidelines for febrile neutropenia are associated with reduced ICU admissions, septic shock, and mortality.^8,16 Centers have reported improved outcomes, including decreased mortality, after introducing best-practice bundles.^17,18 Centers in LMICs may have insufficient resources to address all causes of deviation, particularly inability to afford services. Nevertheless, cancer-directed therapy should not be administered in the absence of a supportive care system adequate to manage complications, including infections. Poor adherence to fever-management algorithms is a marker for suboptimal infection care and should result in the delivered cancer care being limited in its intensity.

Not all deviations affected patients’ length of hospitalization. Failure to appropriately identify and treat high-risk patients and missing ordered antimicrobials resulted in suboptimal therapeutic responses and prolonged hospitalization relative to that in episodes without the indicated deviations. Surprisingly, however, failure to perform blood cultures was not associated with prolonged hospitalization, although the associated variable of pathogen isolation was a significant predictor in both of our models. Blood cultures establish the local pathogen profile and customize individual therapy, thereby avoiding unnecessary antibiotic exposure and potentially shortening hospitalization.^19–21 A low overall blood-positivity rate may have masked this effect, illustrating the need for quality assurance with periodic retraining in collection practices. Nevertheless, as we made considerable efforts to provide adequate training during implementation, provider education alone is clearly insufficient; limiting deviations will require systemic redundancies.

In response to our study findings, we recommend giving nursing staff oversight over order entry and completion, which may improve recognition of incomplete orders. Ward-based nursing staff are more familiar than rotating residents with ward protocols and are also ideal candidates to review risk stratification for accuracy. Locally, we suggest prioritizing nursing education and engagement, as well as providing financial assistance from institutional or non-institutional sources for blood cultures and key antimicrobials. Nevertheless, prioritization at other sites will depend on local deviation patterns and their outcomes.

A universal algorithm for LMICs is not feasible; however, our method and template provide a model for the design, implementation, and evaluation of such algorithms. Combining experiences from sites with different infection profiles and health systems may identify common problems and generate solutions that are relevant across LMICs. For example, our results corroborate published reports that Gram-negative organisms are more frequent in LMICs,^1,22 which implies that procuring vancomycin or other agents targeting resistant Gram-positive organisms may not be a top priority in those countries. Conversely, the relatively frequent isolation of ESBL-producing E. coli mirrors reports of an increased burden of resistant Gram-negative organisms in LMICs.²³ In many LMICs, this is addressed by ordering an aminoglycoside, but drug levels are not routinely checked and toxicities and resistance can develop. Carbapenems are the treatment of choice for serious infections by ESBL-producing organisms; however, these drugs may be prohibitively expensive in LMICs. No patient infected with an ESBL-producing organism in this study was treated with a carbapenem, but none of those infected patients was subsequently readmitted with a recrudescent infection. These findings underscore the importance of tailoring antimicrobial therapy rather than following guidelines that have not been locally validated.

Additional areas requiring local validation include risk-stratification approaches.⁵ Although certain predictors derived from HICs, such as nonrespiratory focus and pathogen detection, had reproducible relevance in Davao, clinical criteria require further study in LMICs, as treatment protocols and comorbid conditions differ. Furthermore, some laboratory tests used for stratification are unavailable or prohibitively expensive in LMICs. Our simplified risk-stratification approach used clinical criteria but no laboratory studies beyond the complete blood count, because obtaining supplementary laboratory tests locally was often not feasible. Cost analysis is required to determine the appropriateness of targeted laboratory testing in LMICs.

Although not generalizable to all LMICs, the local resources and challenges in Davao are probably more widely applicable than are experiences in HICs. When interpreting our results, it should be noted that this population, except for the few patients with AML, received no antibacterial or antifungal prophylaxis. Nor were central venous lines used, which may have decreased the risk of Gram-positive bacterial and/or fungal infection.^22,24 No patient developed pulmonary or disseminated fungal infection, and none received antifungal therapy more aggressive than fluconazole. This may reflect the lower intensity of anti-cancer treatment relative to that in HICs, resulting in less prolonged neutropenia; however, the evaluation is confounded by the limited access to diagnostic studies beyond culture. As access to more aggressive cancer treatments, antimicrobials, diagnostic studies, and central venous line placement increases, the burden of fungal infection in pediatric oncology patients in LMICs will merit further investigation.

Although our study was designed to mitigate threats to data collection and validity, some limitations were unavoidable. The prospective design enabled us to obtain reasons for deviation; however, because those responses were not anonymous, they were subject to provider recall bias. The low number of complications precluded our demonstrating associations between deviation and mortality or ICU admission. Having anticipated this, we used the length of hospitalization as a proxy marker for clinical status by assuming that all patients in the study remained hospitalized for reasons related to febrile neutropenia. This assumption may be invalid. As we could not discern when a patient would have been discharged had they not needed chemotherapy, and because receipt of chemotherapy was not an explicitly collected variable, our analysis did not attempt to account for stays due to chemotherapy. The greatest limitation, however, may be our inability to quantify changes in outcomes between the pre- and post-algorithm eras because of the lack of reliable baseline data. As developing a fever-management algorithm was deemed part of routine practice improvement, deferring its implementation to collect baseline data was judged unethical. We believe, however, that the existing literature is sufficient to establish the benefits of guideline use.

Tailored algorithms for fever management in pediatric oncology patients can be developed for LMICs; however, implementing such algorithms faces practical obstacles. Decreasing selected algorithm deviations may shorten hospitalization and maximize the use of limited resources, thereby improving the quality of care. Operational research involved in generating and evaluating fever algorithms can direct resource prioritization at individual sites but, more importantly, could prove invaluable in constructing best practices for supportive care of children with cancer in LMICs.

Supplementary Material

Supplementary Figure S1:

The resource-tailored algorithm used for adherence evaluation. Steps for which adherence was measured are indicated with a box. Abbreviations: T, temperature; hr, hour; CBC with diff, complete blood count with differential; f/u, follow up; Cx, culture; CXR, chest x-ray; resp S&S, respiratory signs and symptoms; STAT, immediately; CCBDU, Children’s Cancer and Blood Diseases Unit; Mero, meropenem; CBC & diff q2ds, CBC with differential every 2 days; BdCx, blood culture; ANC, absolute neutrophil count; PO, by mouth; Clin, clinically; D/C, discharge; Vanco, vancomycin; Clinda, clindamycin; Abx, antibiotics; Cipro, ciprofloxacin; Δ, Change; w/, with.

Supplementary Table S1:

Frequency of reasons why blood culture was not obtained

Supplementary Table S2:

Deviations occurring by day

Supplementary Table S3:

Frequency of symptoms reported on admission (N = 141 febrile episodes)

Supplementary Table S4:

Factors associated with duration of hospitalization

Abbreviations

ANC

absolute neutrophil count

HIC

high-income countries

ESBL

Extended-spectrum beta-lactamase

LMIC

low- and middle-income country

PCC

pediatric cancer center

St. Jude

St. Jude Children’s Research Hospital