Cross-sectional analysis: clinical presentation of children with persistently low ALP levels

Study design and sampling frame

The non-interventional retrospective ALPHA study was sponsored by the pharmaceutical company Alexion Pharma Germany GmbH. To obtain a sufficiently large set of data from children presenting with persistently low ALP levels in paediatric clinics in Germany a multicentre approach was adopted. This cross-sectional study collected data from patient laboratory and medical records over six months in 2016–2017 at nine German specialised centres (Supplementary Table 1A). Laboratory records were screened to identify patients with repeated readings of low ALP activity over five years prior to the start of the study. Subsequently, a patient chart review was performed. The number of datasets to be included per centre was restricted to a maximum of 100 with no minimum number of records per centre (Supplementary Table 1B).

Records for patients <18 years of age at the time of abnormal ALP value detection with ALP readings below the lower reference limit were eligible. ALP enzyme activity was defined as low based on age-and sex-adjusted local standards. If none was available, a cut-off value of 100 U/L until the age of 12 years for girls and 14 years for boys and in older adolescents of 50 U/L for girls and 70 U/L for boys were used. Thereby, a safety margin with respect to the reference range was introduced, to ensure that all included ALP readings were below the established lower limit of normal [8, 10, 17]. Data were excluded if patients (1) had at least one ALP assessment within the age- and sex-adjusted normal range by history, (2) had a diagnosis of an oncological or a haematological disorder or (3) were in need of intensive care at the time of ALP assessment.

Data collection

Demographic data, biochemical data and clinical data, such as diagnosis, signs and symptoms and information about medication at the time of low ALP assessment, were collected. The data were entered at the centre into an electronic database, without links to any personal subject identifiers like name or date of birth. To ensure anonymization, a randomised centre ID was allocated to each participating site by an independent third party, so that neither the sponsor nor the evaluators could match a dataset to an individual patient or to a participating site. According to applicable German law, a scientific analysis based on a mere collection of anonymised data from routine clinical records does not require approval by a review board. Approval of the institutional ethics committees was therefore waived, and informed consent was not required.

The following variables were extracted from patient records and entered into a GCP-compliant electronic data capture system (MARVIN, xclinical, Munich, Germany):

1. Demographic data: Age (years), sex (male, female), height (cm) and weight (kg) at the time of the most recent ALP value.

2. Biochemical data: Most recent serum ALP values in U/L. Calcium and phosphate serum levels are classified as “normal”, “low” or “high”.

3. HPP: A diagnosis of HPP at the time of the most recent ALP measurement was indicated as “yes” or “no”. A diagnosis of HPP was established according to the routine procedure of the respective centres. No specific diagnostic criteria were defined per study protocol.

4. Other diagnoses and medications: Other diagnoses and medications could be entered as free text.

5. Signs and symptoms: Presence of the following symptoms typical for HPP was specified: short stature, muscle weakness, delayed motor development, gait abnormalities, bone deformities, hypomineralisation, seizures, muscle/bone/joint/lower extremity pain, respiratory disorder, high phosphate or high calcium. All were assessed as “yes” or “no”.

Data evaluation

To adjust for sex and age, z-scores for height, weight and ALP values were determined [10, 18]. ALP z-scores were calculated using ALP percentile reference data from Zierk et al. [10]. Percentiles of observed ALP values were calculated (with interpolation) using age and gender-matched percentile reference data, and z-scores were calculated based on the determined ALP percentiles for observed ALP values. ALP level of zero was used as zero percentile for all ages to calculate percentiles of ALP values that are below the 1st percentile. For the calculation of height and weight z-scores (Z), the formula for a given physical measurement X (weight or length) is: Z = ( ( X M ) L ) − 1 L S , L ≠ 0  or  Z = ln ( X M ) S , L = 0 , where L, M and S are the estimated parameters from the CDC growth chart reference data corresponding to the age [18]. Height and weight z-scores above 4 or below −4 were assessed for plausibility, and non-plausible values were excluded.

For further analyses, data sets of the total cohort (Group 1) and the following subgroups were considered: data sets with diagnoses other than HPP or unclear diagnoses (Group 2), and datasets with a documented diagnosis of HPP (Group 3). Incidences of signs and symptoms typical for HPP and the number of concomitant typical symptoms per patient were calculated in the subgroups of patients without (Group 2) and with (Group 3) documented HPP.

All diagnoses other than HPP as well as medications that were entered into the database were screened for the presence of one or more known conditions associated with low serum ALP values. The following conditions were considered relevant: hypothyroidism, calorie depletion, zinc deficiency, magnesium deficiency, vitamin B12, C or D deficiency, previous major surgery or trauma, severe anaemia, celiac disease, growth hormone deficiency, impaired kidney or liver function, Wilson’s disease, abnormal serum parathyroid hormone or ceruloplasmin levels, as well as the administration of contraceptives/oestrogen or steroids [2, 19], [20], [21], [22], [23], [24], [25], [26]. If multiple entries were documented for a single patient, all were considered. The respective incidences were calculated for the total cohort.

To assess whether there is a correlation between low serum ALP levels and body height, a potential relationship was evaluated using calculated z-scores for both variables.

Medians and interquartile ranges (IQRs), as well as 25th and 75th percentiles, were calculated for continuous variables. Proportions were determined for categorical parameters. Linear regression to study the relationship between z-scores of body height and ALP values was calculated based on the Least Squares method. A two-sided F-test with a cut-off value for significance of p<0.5 was applied. The analyses were done using R software (version 3.6.3, R Foundation for Statistical Computing, Vienna, Austria).

Bias was reduced by using a multicentre approach and a random sampling of records provided by the laboratories, including multiple TNSLAP readings when available. As the study objectives were explorative, formal sample size estimation was not required.

Study cohort

Data for 393 patients from the nine participating centres were entered into the database. As data were fully anonymised, the patient IDs could not be allocated to the respective centres but only to the centre ID. Records of 23 patients were excluded from the data set as per the exclusion criteria, resulting in 370 data sets for further analysis (Figure 1). Additionally, three and nine patients were excluded from the evaluation related to height and weight based on non-plausible z-score values of above 4 or below −4, respectively. Missing data for specific variables resulted in a smaller group size available for the respective analysis.

Figure 1:

Flowchart depicting disposition of patient database entries and assignment to the analysis groups.

Out of 370 patients (Group 1), 15 patients had a confirmed diagnosis of HPP (Group 3), whereas the subgroup without a diagnosis of HPP (Group 2) comprised 355 patients. Due to the strongly inhomogeneous sample sizes and patient age distribution of the subgroups, no inferential statistics were deemed feasible. Therefore, only the results of the descriptive statistics are presented. The demographic data are summarised in Table 1.

ALP values

The median ALP activity in the total cohort was 62.0 U/L (IQR = 42.0 U/L) and the respective z-score was −2.07 (IQR = 0.87). The ALP serum levels were on average lower in the group with established HPP diagnoses (Group 3) compared to the Group 2 with medians of 52.0 U/L (IQR = 44.0 U/L) vs. 64.0 U/L (IQR = 47.9 U/L). The median ALP z-score values were −2.52 (IQR = 0.20) and −1.96 (IQR = 0.87), accordingly.

Diagnoses and medications associated with low ALP levels in the core group

For 241 (67.9%) patients without a confirmed HPP diagnosis (Group 2), one or more medical conditions were specified, which are known to be associated with low ALP serum levels other than HPP. Table 2 provides the respective incidences. In the data sets of 54 (15.2%) patients, the diagnosis was stated as not completed. For the remaining 60 (16.9%) patients there was insufficient information to allocate them to the listed specific conditions.

Signs and symptoms

The pattern of signs and symptoms overlapped between the core group and the group with documented HPP, and patients with clinical features consistent with HPP could be found in both groups. As expected, relevant symptoms were relatively more frequent and more clustered in the HPP group. Six of the HPP patients had two or more characteristic symptoms (Figure 2A). Short stature was the most prevalent feature in both subgroups (Figure 2B).

Figure 2:

(A) Accumulation of typical signs and symptoms in individual patients. (B) Percentage of patients presenting with symptoms, which are typically associated with HPP in the group of patients without (Group 2) and with documented HPP diagnosis (Group 3).

Relationship between body height and ALP levels

Interestingly, a weak statistically significant positive linear relationship between z-scores of ALP levels and the body height was identified (Figure 3):

Figure 3:

Relationship between body height and ALP values (as z-scores) in the total cohort (Group 1; p=5.83e-8).

Study limitations

The main limitation of the study is its retrospective character. As it was cross-sectional data analysis, no causal relationships can be deduced. Therefore the origin of the low ALP readings in many patients remains unclear. In addition, data quality is heterogeneous, since the information provided from the patient records was partially incomplete and the diagnostic procedure used for establishing the HPP diagnosis was not assessed. Further, it was not part of the study to follow-up the patients based on their records. Therefore, it is possible that within the group of low ALP readings, cases of undiagnosed HPP may be included, which would result in a partial overlap of the subgroups. While the sampling frame was not rigorously defined, we consider a random selection from available laboratory records at nine specialised centres throughout Germany a valid database. Bias was reduced by clearly predefined in- and exclusion criteria, a GCP-compliant data management and a multicentre approach in an academic setting.