Serum Phosphorus and Hypophosphatemia During Therapy of Diabetic Ketoacidosis in Children: Single-Center, Retrospective Cohort 2016–2022

Abstract

OBJECTIVES:

To assess factors associated with serum phosphorus (P) and hypophosphatemia in children with type 1 diabetes mellitus (T1DM) treated for diabetic ketoacidosis (DKA).

DESIGN:

Retrospective cohort.

SETTING:

Community-based PICU in a university-affiliated hospital.

PATIENTS:

Patients 1–20 years old with T1DM hospitalized for DKA from July 1, 2016, to July 31, 2022.

INTERVENTIONS:

None.

MEASUREMENTS AND MAIN RESULTS:

We collected age, sex, duration of T1DM, conscious state at presentation, and most recent glycohemoglobin level. P was tested initially and then every 4 hours. Probability of hypophosphatemia and time to hypophosphatemia and hospital length of stay (LOS) were analyzed via binomial and linear mixed-effects regression analyses, respectively. A total of 852 DKA episodes occurred in 365 patients (46.3% female, median age 14.7 yr), of which 158 (18.5%) episodes were new-onset T1DM. Hypophosphatemia developed during 656 of 852 (77%) episodes, including 49 of 852 (5.8%) episodes of severe hypophosphatemia with median (interquartile range) onset 8.0 hours (4.7–11.9 hr) and 12.0 hours (8.1–17.6 hr), respectively, following initiation of therapy. Higher glycohemoglobin was associated with greater odds of hypophosphatemia (odds ratio [OR], 1.22; p < 0.001). However, lower odds of hypophosphatemia were associated with older age (OR, 0.89; p < 0.01), male (OR, 0.11; p = 0.01), longer T1DM duration (OR, 0.87; p < 0.001), and having initial normal conscious state (OR, 0.18; p < 0.01). Older age (3.0%/yr; p = 0.02), T1DM duration (4.1%/yr; p = 0.01), and initial serum P (23.4%/mg/dL; p < 0.001) were associated with later hypophosphatemia. LOS was shorter with increased T1DM duration (3.6%/yr; p < 0.001) and normal conscious state (33.1% shorter; p < 0.001), but longer with increasing glycohemoglobin (4.0%; p < 0.001). All patients survived with normal neurologic function.

CONCLUSIONS:

Higher glycohemoglobin was associated with greater odds of hypophosphatemia and longer LOS. Older male, longer duration of T1DM, and conscious at admission were factors associated with lower odds of developing hypophosphatemia and with later onset when it occurred. Hypophosphatemia was associated with longer LOS.

RESEARCH IN CONTEXT.

• We need to learn about the frequency of and the timeline for occurrence of hypophosphatemia in children with type 1 diabetes mellitus (T1DM) who have diabetic ketoacidosis (DKA).

• This study provides data on the pattern of changes in serum phosphorus (P) and the severity of hypophosphatemia during therapy for DKA in children with T1DM.

WHAT THIS STUDY MEANS.

• In our 2016–2022 cohort of T1DM patients with DKA, 77% of developed hypophosphatemia at a median of 8 hours after starting therapy. Severe hypophosphatemia occurred in 5.8% of episodes and developed 12 hours after starting therapy.

• Higher glycohemoglobin at presentation was associated with greater odds of developing hypophosphatemia and had longer length of stay. In contrast, older, male, longer T1DM duration, and being conscious at presentation were associated with lower odds of developing hypophosphatemia.

• During DKA treatment, severe hypophosphatemia is treatable and develops several hours after starting therapy. However, the associated explanatory factors at presentation are not useful for targeting those with greater odds of severe hypophosphatemia. Generalized monitoring is therefore still needed.

Therapy for diabetic ketoacidosis (DKA) in children includes administration of IV fluid and insulin infusion (1–3). This combination of therapies can lead to progressive decline in serum phosphorus (P) that may result in hypophosphatemia (1, 4). Serum P is frequently measured in critically ill children with DKA to monitor alterations that may require intervention (1, 2, 4), and normative data have been established in children (1, 5, 6). National/International guidelines (1, 6, 7) on monitoring of serum P are based mostly on expert opinions and do not provide a timeline on when clinicians should expect the development of hypophosphatemia, and there are limited data on the pattern of changes in serum P during DKA therapy in children (4). Therefore, the purpose of this study was to assess dynamics of serum P in pediatric patients with type 1 diabetes mellitus (T1DM) during DKA therapy and to evaluate factors associated with hypophosphatemia, time to hypophosphatemia development, and hospital length of stay (LOS).

MATERIALS AND METHODS

Consecutive pediatric patients with T1DM who were initially evaluated and treated in the emergency department (ED) for DKA and subsequently admitted to the PICU from July 1, 2016, to July 31, 2022, were enrolled. The Hurley Medical Center Institutional Review Board approved the study (Title: Hypophosphatemia in children with diabetic ketoacidosis; Approval Number: 2016179-1; Approval Date: March 14, 2023) and waived the need for consent due to the retrospective nature of the study. Procedures were followed in accordance with ethical standards of the responsible committee on human experimentation at the hospital and with the Helsinki Declaration of 1975.

Patients 1–20 years old with new onset or previously diagnosed T1DM with DKA were identified through Electronic, Process, Instrumentation, and Control Systems electronic health records (EPIC Systems Corporations, Verona, WI) using International Classification of Diseases, 9th Edition codes for DKA in children hospitalized during 2016–2022.

Definition of DKA and Standard Management of DKA

Treatment for DKA included initial IV fluid resuscitation followed by IV fluid administration using the two-bag system as previously described (9) and based on the Holliday and Segar (10) formula. Severe hypophosphatemia was treated with an infusion of potassium phosphate (0.5 mmol/kg body weight with a maximum dose of 15 mmol in 100 mL of 0.9% normal saline [NS]) delivered IV over 4 hours. Hypokalemia (serum potassium < 3.5 mEq/L) was treated with infusion of potassium chloride (0.5 mEq/kg body weight with a maximum dose of 10 mEq in 100 mL of 0.9% NS infused over 1 hr) and repeated until serum potassium increased to greater than or equal to 3.5 mEq/L. All patients received an IV infusion of crystalline insulin at 0.1 U/kg/hr.

The two-bag system of IV fluid and insulin infusion were continued until DKA resolved and the patient was transitioned to subcutaneous insulin as previously described (9).

Data Collection

Baseline characteristics included age, sex, T1DM duration, number of episodes of DKA requiring hospitalization, and conscious state at the time of initial presentation to the ED, defined as the ability to open eyes in response to a command. Results of the biochemical profile on the blood sample collected at initial presentation were documented.

The most recent glycosylated hemoglobin level was documented, and serum glucose was measured hourly. Serum electrolytes were measured at presentation to the ED and then every 2 hours, while serum P, magnesium, and venous blood gases were measured every 4 hours until DKA resolved.

The number of patients experiencing hypophosphatemia or severe hypophosphatemia, time to hypophosphatemia, nadir serum P (nadir P), and hospital LOS were recorded.

Statistical Analyses

Detailed statistical methods are available in the STROBE checklist and Supplemental Methods, http://links.lww.com/PCC/C560). Analysis was conducted using R (Version 4.2.2), with assistance from “tidyverse” package (Version 1.3.2), within RStudio (Version 2023.06.0, build 421; Posit PBC, Boston, MA).

Explanatory variables for hypophosphatemia development were evaluated using a mixed effects binomial generalized linear model. Time to hypophosphatemia and LOS were analyzed using linear mixed models, with time logarithmically transformed before analysis. For each model, random intercepts were estimated for each patient, and fixed effects included age, sex, glycohemoglobin, conscious state, initial serum P, and T1DM duration. Time to hypophosphatemia was included as an additional fixed effect in LOS analyses. Similar analyses were performed for severe hypophosphatemia but using fixed effects models due to the small number of patients who experienced multiple severe hypophosphatemia episodes.

Data are summarized as frequencies, proportions, and percentages. Distributions are summarized with median (interquartile range [IQR]). Associations are presented as odds ratios (ORs) with 95% CI. For all analyses, p values of less than 0.05 were considered statistically significant. Given the exploratory nature of the analyses, we did not correct for multiple hypothesis tests.

RESULTS

A total of 365 patients (46.3% female, with median age 14.7 yr [IQR, 12.2–17.2 yr]) experienced 852 episodes of DKA, of which 18.5% (158/852) were newly diagnosed T1DM episodes. The median T1DM duration was 3.5 years (IQR, 1.1–5.3 yr) for patients with an established diagnosis of T1DM. Overall, 759 of 852 patients (89%) were conscious at presentation. One patient younger than 1 year old was excluded from subsequent analyses.

The initial median biochemical parameters of all 852 episodes were as follows: pH 7.2 (IQR, 7.1–7.3), blood glucose 467 mg/dL (IQR, 345–600 mg/dL), sodium 135 mEq/L (IQR, 132–138 mEq/L), potassium 4.9 mEq/L (IQR, 4.4–5.4 mEq/L), bicarbonate 11.0 mEq/L (IQR, 7.0–16.0 mEq/L), blood urea nitrogen 18.0 mg/dL (IQR, 14.0–23.0 mg/dL), creatinine 1.1 mg/dL (IQR, 0.9–1.4 mg/dL), and serum magnesium 2.1 mg/dL (IQR, 1.9–2.4 mg/dL) (Supplemental Table 1, http://links.lww.com/PCC/C560). The most recent median glycohemoglobin measurement was 12.1% (IQR, 10.7–13.7%), with 40 of 851 episodes (4.7%) associated with normal glycohemoglobin.

Regarding serum P level, the median was 4.7 mg/dL (IQR, 3.7–6.0 mg/dL). Overall, there were 656 of 851 (77%) episodes subsequently complicated by hypophosphatemia, of which 49 of 656 (7.5%) were in the severe category; alternatively, 49 of 851 (5.8%). The severe hypophosphatemia episodes were treated with an infusion of potassium phosphate without documentation of clinical harm. All patients survived. None of the patients developed seizures, or required interventions for cardiac arrhythmias, or needed respiratory support.

Factors Associated With Hypophosphatemia Development

DKA episode data at presentation with characteristics in relation hypophosphatemia are shown in Table 1. Higher glycohemoglobin was associated with greater odds of subsequently developing hypophosphatemia (OR, 1.22 [95% CI, 1.11–1.34]; p < 0.001). Older age (OR, 0.89/yr [95% CI, 0.81–0.98]; p < 0.01), being male (OR, 0.11 [95% CI, 0.02–0.57]; p = 0.01), longer T1DM duration (OR, 0.87/yr [95% CI, 0.80–0.94]; p < 0.001), and normal conscious state at initial presentation (OR, 0.18 [95% CI, 0.05–0.62]; p < 0.01) were each associated with lower odds of developing hypophosphatemia.

TABLE 1.

Factors Associated With Hypophosphatemia Episodes

Variable	Hypophosphatemia (n of 803 Episodes)			Severe Hypophosphatemia (n of 49 Episodes)
	Estimate ± sem	z Value	p	Estimate ± sem	z Value	p
Intercept	3.37 ± 1.14	2.96	0.00b	–1.77 ± 1.58	–1.12	0.26
Age (yr)	–0.12 ± 0.05	–2.47	0.01c	0.10 ± 0.07	1.53	0.13
Sex (males)	–2.25 ± 0.86	–2.62	0.01b	–1.05 ± 1.53	–0.69	0.49
Age × sex	0.09 ± 0.06	1.54	0.12	0.04 ± 0.10	0.40	0.69
Duration of type 1 diabetes mellitus (yr)	–0.14 ± 0.04	–3.43	0.00a	–0.17 ± 0.08	–2.14	0.03c
Last glycohemoglobin	0.20 ± 0.05	4.22	0.00a	0.16 ± 0.08	1.93	0.05
Conscious state (normal)	–1.72 ± 0.63	–2.71	0.01b	–2.11 ± 0.43	–4.88	0.00a
Initial serum phosphorus	–0.05 ± 0.05	–0.99	0.32	–0.49 ± 0.13	–3.92	0.00a

^ap < 0.001.

^bp < 0.01.

^cp < 0.05, p < 0.10.

Higher glycohemoglobin was associated with greater odds of developing severe hypophosphatemia (OR, 1.17 [95% CI, 0.3–1.37]; p = 0.05). In contrast, longer T1DM duration (OR, 0.84/yr [95% CI, 0.72–0.98]; p = 0.03), normal conscious state at initial presentation (OR, 0.12 [95% CI, 0.05–0.29]; p < 0.001), and higher initial serum P level (OR, 0.61/mg/dL [95% CI, 0.47–0.77]; p < 0.001) were each associated with lower odds of severe hypophosphatemia.

Time to Hypophosphatemia

Older age (3.0%/yr [95% CI, 1.0–5.0%]; p = 0.02; Fig. 1A), T1DM duration (4.1%/yr [95% CI, 0.1–8.2%]; p = 0.01; Fig. 1B), and initial serum P (23.4%/mg/dL [95% CI, 18.6–28.3%]; p < 0.001) were associated with later onset of hypophosphatemia. We failed to identify an association between being male rather than female and experiencing subsequent hypophosphatemia (49% [95% CI, –6.8% to 138.8%]; p = 0.10) (Fig. 1A). However, the point estimate and 95% CI shows that we cannot exclude the possibility that, on average, males may have up to a 90% later development of hypophosphatemia.

Figure 1.

Time to hypophosphatemia across several variables. A, Age and sex. Blue dots (male) and red asterisks (female) represent individual visits. Blue dotted line (male), red dashed line (female), and black line (all) represent the linear trend of time to hypophosphatemia against age. B, History of diagnosis with diabetes. Dots represent individual visits; black line represents linear trend of time to hypophosphatemia against years diagnosed with diabetes. C, Glycohemoglobin (A1c) level. Dots represent individual visits; black line represents linear trend of time to hypophosphatemia against most recent A1c. Error bars represent 95% CIs, weighted by abundance of data. T1DM = type 1 diabetes mellitus.

There was a 1.0% (95% CI, 0.8–1.2%; p < 0.01) decline in serum P for every hour past the initiation of therapy for DKA. Delayed onset of severe hypophosphatemia was associated with higher initial serum P levels (35.0%/mg/dL [95% CI, 17.7–54.8%]; p < 0.001).

When initial serum P level was less than 5 mg/dL, there was a near-linear relationship, with higher initial serum P associated with a higher nadir P. When initial serum P was greater than 5 mg/dL, the relationship approached a quadratic term, such that a higher initial serum P was associated with a lower nadir P level (Fig. 2). Higher initial serum P levels was associated with 32.3% greater nadir P levels for the linear term (95% CI, 24.8–40.3%; p < 0.01), but 2.0% lower nadir P levels for the quadratic term (95% CI, 1.6–2.4%; p < 0.01).

Figure 2.

Relation between nadir serum phosphorus (P) level (X-axis) and initial serum phosphorus (iP) level (Y-axis). Blue line (median nadir P level across iP levels). Light blue ribbons (5–95% percentiles); dark blue ribbons (25–75% percentiles). Median line plotted with locally estimated scatterplot smoothing function. Values of ribbons were binned by rounding iP levels to nearest integer. Markers are displayed with jitter to avoid overplotting.

Hospital LOS

The overall median hospital LOS was 61.1 hours (IQR, 46–79 hr) (Fig. 3). There was an association between shorter LOS and increasing T1DM duration (3.6%/yr [95% CI, 2.1–5.0]; p < 0.001) and between longer LOS and increasing glycohemoglobin (4.0% [95% CI, 2.2–5.5%]; p < 0.001) (Table 2 and Fig. 3, B and C). Normal conscious state was associated with 33.1% shorter LOS (95% CI, 23.9–41.1%; p < 0.001) (Table 2). Never developing hypophosphatemia was also associated shorter LOS. There was a significant age-by-sex interaction that showed that on a logarithmic scale, there was an associated decrease in LOS with age in males but not females (Fig. 3A).

Figure 3.

Hospital length of stay (LOS) across several variables. A, Age and sex. Blue dots (male) and red asterisks (female) represent individual visits. Blue dotted line (male), red dashed line (female), and black line (all) represent the linear trend of log (LOS) against age. B, History of diabetes. Dots represent individual visits; black line represents linear trend of log (LOS) against years diagnosed with diabetes. C, Glycohemoglobin (A1c) levels. Dots represent individual visits; black line represents linear trend of log (LOS) against most recent A1c. Error bars represent 95% CI, weighted by abundance of data. T1DM = type 1 diabetes mellitus.

TABLE 2.

Factors Associated With Hospital Length of Stay

Variable	Estimate ± sem	Degrees of Freedom	t Value	p
Intercept	4.04 ± 0.16	623	25.18	0.00a
Age (yr)	0.00 ± 0.01	347	0.55	0.58
Sex (male)	0.21 ± 0.13	408	1.69	0.09
Age × sex	–0.02 ± 0.01	383	–1.80	0.07
Duration of type 1 diabetes mellitus (yr)	–0.04 ± 0.01	425	–4.77	0.00a
Last glycohemoglobin	0.04 ± 0.01	609	4.75	0.00a
Conscious state (normal)	–0.40 ± 0.07	743	–6.10	0.00a
Initial serum phosphorus (mg/dL)	0.01 ± 0.01	764	0.79	0.43

^ap < 0.001.

DISCUSSION

In this single-center, retrospective study (2016–2022), the prevalence of developing hypophosphatemia was 77% of episodes of DKA, with 5.8% of episodes subsequently developing severe hypophosphatemia. Severe hypophosphatemia was typically preceded by hypophosphatemia. These results are consistent with previous reports on the pattern of development of hypophosphatemia in adult patients with DKA (11, 12). DKA produces metabolic acidosis, which is associated with an extracellular shift of P and may contribute to higher serum P levels during the evolution of DKA (2). In addition, volume depletion may lead to impaired renal perfusion with resultant reduced elimination of P (1, 2). These combined factors likely contributed to the pattern of serum P in this cohort. Serum P was at the upper limit of the normal range for age in most patients at the time of presentation with DKA and hypophosphatemia developed after therapy initiation. These observations are consistent with the findings of van der Vaart et al (11) and Shen and Braude (12) in adult patients with DKA.

Insulin infusion rectifies insulinopenia, and accelerates glycolytic pathways that involve phosphorylation, which requires incorporation of inorganic P (6). This is associated with a progressive decline in serum P (6, 7) due to transcellular shift of P. Additionally, volume expansion improves renal perfusion, leading to enhanced urinary P excretion, contributing further to the progressive decline in serum P. These combinations of factors explain the initially normal or high serum P and subsequent decline following initiation of therapy for DKA, which was observed in this study (6, 7, 9).

Older male patients with a longer duration of T1DM and those who were conscious at the time presentation for DKA had lower odds of developing hypophosphatemia, while those with higher glycohemoglobin had greater odds. Sex differences in serum P have been reported, but data continue to accumulate on the underlying mechanism related to this phenomenon (13). Estrogen has been shown to lower serum P and increase fraction of excretion of P. These biological differences may explain the lower odds of developing hypophosphatemia in older male patients in this study compared with female (13, 14).

There were lower odds of developing severe hypophosphatemia when there was an associated increased T1DM duration, increased initial serum P level, and the finding of normal consciousness at initial presentation in the ED. Our results also show that older age, T1DM duration and higher initial serum P were associated with later onset of hypophosphatemia during therapy for DKA. The explanatory variables for severe hypophosphatemia were less obvious. However, it was treatable and occurred at a median of 12 hours following initiation of therapy for DKA. Patients who were conscious at the time of presentation are likely to have had less severe metabolic derangements at the cellular level and thus were less likely to develop fluid and electrolytes abnormalities including hypophosphatemia.

Higher glycohemoglobin levels were associated with longer LOS, while longer T1DM duration and normal conscious state corresponded to shorter LOS. These findings suggest that children with poorer T1DM control are more likely to develop hypophosphatemia and require higher resource utilization. The mechanism for this observation is a matter of conjecture, but one potential mechanism is the process of phosphorylation in these children. Insulin binding triggers autophosphorylation of the insulin receptor on tyrosine residues within its intracellular domain. This leads to activation of tyrosine kinase, which is essential for initiating the downstream signaling cascade that leads to insulin’s metabolic effects, including glucose uptake by cells, glycogen synthesis, and protein synthesis (15). It seems likely that patients with poorly controlled T1DM with higher glycohemoglobin will require more insulin therapy, therefore, more phosphorylation and more depletion of serum P.

Limited studies in adults have documented hypophosphatemia occurrence during DKA therapy (11, 12, 16, 17). Van der Vaart et al (11) retrospectively evaluated serum P in a 2005–2020 cohort of 80 adult patients with T1DM experiencing 127 DKA episodes. Serum P was measured in only 66% (84/127) of DKA episodes. Hypophosphatemia developed in 74% patients (62/84), which is comparable to the frequency of occurrence of hypophosphatemia in our study. The time to lowest serum P was 16 hours, which is longer than the onset of hypophosphatemia in our study; however, serum P was not monitored as frequently compared with our study and it is possible that earlier onset of hypophosphatemia was missed (11).

Shen and Braude (12) retrospectively analyzed P in 64 DKA episodes in a 2007–2010 cohort of 43 adults with T1DM. Overall, 63% of patients had elevated serum P at presentation, and hypophosphatemia was noted in 90% of episodes with a mean ± sd of 22 ± 14 hours following initiation of therapy for DKA. In our study, hypophosphatemia onset occurred earlier following initiation of therapy for DKA; however, Shen and Braude (12) reported a delay in measuring serum P following initiation of therapy for DKA, serum P was not monitored frequently, and details of DKA management were not provided; therefore, a direct comparison to our study is difficult.

In children, Smuel-Zilberberg et al (18) reported a retrospective (2008–2018) cohort of 356 children in Israel with DKA. The authors compared biochemical parameters and time to recovery from DKA between children with new-onset T1DM and children with established T1DM. While it is unclear when serum P was measured during DKA therapy and serum P dynamics were not documented, the authors reported that serum P and potassium levels were significantly lower while serum sodium and chloride levels were higher in patients with newly diagnosed T1DM compared with children with an established diagnosis. The authors did not provide details of serum P levels during DKA in these children.

Our study provides a timeline for the chronology of development of hypophosphatemia and severe hypophosphatemia and factors associated with these alterations. Each additional year of age was associated with an additional 3% delay to the onset of hypophosphatemia and an associated 8% delay to development of severe hypophosphatemia following initiation of DKA therapy. Our study included patients with poorly controlled T1DM with a median glycohemoglobin of 12.5%, which is known to adversely affect kidney function (19, 20). Children with a longer duration of poorly controlled T1DM may have subclinical kidney dysfunction (19–21), which may affect the handling of electrolytes including P. This may contribute to delayed hypophosphatemia onset. Additionally, younger children have a faster metabolic rate (22), which may accelerate glycolytic pathways following resolution of insulinopenia. Furthermore, since the decline in serum P during DKA therapy is a function of both transcellular shift and renal excretion of P, higher serum P at the initiation of DKA therapy may be associated with the observed delay in onset of hypophosphatemia in our study; there was a 35% delay in the onset of hypophosphatemia for every mg/dL of higher initial serum P.

DKA treatment requires frequent changes in electrolytes and IV fluid. This may lead to higher resource utilization and lead longer LOS observed in children with hypophosphatemia in this study. Furthermore, children with poor diabetes control are likely to have subclinical reduction of organ functions that may lead to more difficult DKA management and longer LOS.

The main limitation of this study is its retrospective single-center design. The true nadir of serum P may have been missed because of the intermittent measurement of serum P; therefore, results are not generalizable to other clinical settings.

CONCLUSIONS

In this 2016–2022 retrospective study, hypophosphatemia was common during DKA therapy in children with T1DM. Severe hypophosphatemia was present in around one-in-20 cases and developed several hours into DKA therapy. Higher glycohemoglobin was associated with greater odds of developing hypophosphatemia and longer LOS. In contrast, older age, longer T1DM duration, and being conscious at presentation were associated with lower odds of developing hypophosphatemia. Absence of hypophosphatemia was associated with a shorter LOS. This exploratory work may be a starting point to study the clinical significance of hypophosphatemia in children with DKA.

ACKNOWLEDGMENTS

We acknowledge Superior Medical Experts for drafting, biostatistical, and editorial assistance.

Supplementary Material

pcc-26-e77-s001.docx ^{(37.8KB, docx)}