Introduction

Hyponatremia is the most common electrolyte abnormality encountered in health care [1]. It is seen in a wide variety of chronic medical conditions, such as heart, renal, and liver failure, as well as in acute conditions, such as diarrhea and vomiting [2]. Hyponatremia can also be iatrogenic, caused by the intake of common drugs, e.g., thiazide diuretics and antidepressants [3–5]. Severe complications of hyponatremia include coma, permanent brain damage, respiratory arrest, brain stem herniation, and death [6,7]. Studies show that at least 30% of patients in medical, surgical, and psychiatric wards are hyponatremic [8]. One third of all hyponatremia cases result from syndrome of inappropriate anti-diuretic hormone secretion (SIADH) [9]. Advanced age, female sex and low body mass are important risk factors for developing hyponatremia [7,10].

Validity refers to how well a measure assesses what it is intended to measure [11]. Consequently, an evaluation of diagnosis codes in medical records is primarily conducted as quality control. The purpose here is to determine if the principal diagnosis is accurately chosen and reliable [12,13].

Hospitalization is a clinically significant event, often indicating severe disease and it also entails a large cost for society. The principal diagnosis, reflecting the main reason for hospitalization, may therefore be a clinically relevant outcome for register-based studies.

Principal diagnoses have in fact previously been used to investigate the association between drug groups and hospitalization due to hyponatremia [4,14–21], investigating mortality after being hospitalized due to hyponatremia [7] and to which extent clinicians discontinue thiazides after being hospitalized due to hyponatremia [5]. However, these studies rely on the quality of the principal diagnoses: in essence, their validity.

The aim of this study was to assess the validity of ICD-10 (International Classification of Diseases) codes for hyponatremia as principal diagnoses in patients discharged after hospitalization.

Materials and methods

Results

We included in total 428 patients from Södersjukhuset (n = 200), Karlstad Hospital (n = 109) and Falu Hospital (n = 119).

The range of sodium concentrations at admission ranged from < 100 to 139 mmol/L (the patient with sodium of 139 mmol/L, had lower levels when were treated earlier at another hospital), and the lowest levels during hospitalization between < 100 and 135 mmol/L, respectively. The mean (SD) sodium concentration (first) was 120 (±6) mmol/L and the lowest 115 (±8) mmol/L. The mean (SD) of glucose concentrations used to correct the sodium values was 7 (±3) mmol/L. Patients with severe (sodium < 125 mmol/L) and very severe (sodium < 115 mmol/L) hyponatremia, were 54% (231/425) and 31% (130/425) of the patients, respectively.

The median age was 73 (IQR 64–81) and, 69% were women. The mean caretime at the hospital was 5 days. Of the patients 27% were treated in ICU/HDU due to the severe hyponatremia concentrations and/or related comorbidities. The mean (SD) sodium concentration at admission if hospitalized in ICU/HDU was 113 (±6) mmol/L. The most frequent comorbidity was hypertension (n = 284) followed by chronic obstructive pulmonary disease/asthma, diabetes, malignant disease, previous history of a cerebrovascular incident, ischemic heart disease and heart failure (Table 1). Each patient consumed regularly on average 7 prescribed drugs at admission.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Patient characteristics among the study population (n = 428).

https://doi.org/10.1371/journal.pone.0318927.t001

After reviewing the journals, we found that 411/428 of the principal diagnosis were correct resulting in a positive predictive value of 96% (CI 95% 94–98). The characteristics of the 17 false positive diagnoses are shown in Table 2. According to the current validation, the diagnosis most erroneously misplaced for hyponatremia were renal failure, heart failure and urinary tract infections.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. The final diagnoses of the 17 false positive diagnoses with corresponding sodium values.

https://doi.org/10.1371/journal.pone.0318927.t002

Discussion

This retrospective validation study is the first of its kind, focusing on the evaluation of hyponatremia as a principal diagnosis. The results showed that 96% of patients discharged with hyponatremia as a principal diagnosis received the correct diagnosis.

There is a need for a better understanding of the underlying conditions of hyponatremia. One important step towards this goal is the efficient use of population-based registers, which in turn demands reliable outcomes. Hyponatremia as the principal diagnosis has a large potential in this regard. However, a prerequisite for its use is that it is valid, in essence that it is in fact measuring what it is supposed to measure which is the purpose of the current paper. ICD-codes have previously been used as a proxy for hyponatremia in register-based outcome studies. However, evidence shows that the validity of ICD-codes to subnormal sodium concentrations is modest. Holland-Bill et al included patients discharged with any hyponatremia diagnosis, not only principal diagnosis, and showed that ICD-10 codes for hyponatremia in the Danish National Registry of Patients have high specificity (≥99.8%) but very low sensitivity (1.8%) [23]. A Canadian study was restricted to elderly patients, 66 years or older, and to serum sodium concentrations at the time of emergency department contact or at hospital admission. That study showed that the hyponatremia ICD-10 code differentiates between two patient groups with measurements of serum sodium at the emergency department and at hospital admission. However, these codes underestimated the true incidence of hyponatremia due to low sensitivity (7.5% at presentation to the emergency department and 10.6% at hospital admission). The positive predictive value for the emergency department was 96.4% and 82.3% for hospital admission. Distinctions between principal and secondary diagnosis was not considered [24]. Finally, Movig et al observed that ICD codes for hyponatremia at a single hospital were highly specific but demonstrated low sensitivity. The sensitivity of hyponatremia ICD codes was not higher than around 30% for patients with severe hyponatremia (sodium ≤ 115 mmol/L) and corresponding specificities were very high (>99%) [25].

Although sodium values as well as ICD-codes as a surrogate for sodium concentrations may be useful outcomes in research, they may not always be clinically relevant. Though the clinical significance of secondary diagnosis may vary the main reason for hospitalization is important representing a condition associated with more severe disease that demands the resources associated with inpatient care.

Since the review was limited to patients diagnosed with hyponatremia, we could only estimate the positive predictive value (PPV). A more detailed and complete validation including the remaining measures of validity, i.e., negative predictive value (NPV), sensitivity and specificity, was not feasible because it would have required a review of all discharge summaries discharge summaries at Södersjukhuset, Karlstad- and Falu Hospitals during the full study period. With a total catchment area of 1,275,000 individuals the total number of inpatient medical files in need for review would have amounted to several tens of thousands. Unfortunately, this approach was therefore not feasible. Hence, the current study cannot rule out the possibility of a certain proportion of hospitalizations due to hyponatremia were wrongly diagnosed (false negative assessments). However, at the point of discharge from inpatient care, the physician responsible for the inpatient care is obliged to judge the number one main reason for hospitalization. In contrast, hospitalizations with one or several measurements of sodium below the reference interval in a patient where the main reason for care is derived from a totally different area such as an acute cardiovascular disease or infection may not to a similar extent led to a diagnosis of hyponatremia.

A strength of the current study is that we included three hospitals of different sizes, from different parts of Sweden. The PPV for Södersjukhuset, Karlstad- and Falu Hospital respectively were approximately 96% each, similar to the overall PPV. The participation of additional hospitals would of course increase the generalizability of the results further. To avoid/reduce systematical errors related to interpretation we had three independent reviewers. All with several years of clinical practice within the area of internal medicine which increased the precision of the assessments.

We cannot exclude that the true PPV may deviate from the results in the current study. However, when planning the study, a power analysis was performed. Based on the presumptions on a true proportion of correct diagnosis of 0.9, a sample size of 414 could result in a 95% confidence interval with a total width of 0.06.

Conclusion

The current study shows that the positive predictive value for hyponatremia used as a principal diagnosis was 96%. Thus, hospitalization due to hyponatremia appears to be a useful measure in epidemiological studies aiming to target a clinically relevant consequence of hyponatremia.

Supporting information