A total of 1631 hospital discharge reports were collected from the Hospital del Mar Nephrology Service between 2016 and 2018. The documents were anonymized in PDF format from a computer located at Hospital del Mar, where a Bismart programmer had physical access without internet connection to proceed with the elimination of the headers containing patients affiliation data through an automated process developed by Bismart (Barcelona) using Python language, being able to only identify the gender, necessary for the subsequent application of glomerular filtration estimation formulas. Once the headers of the medical discharge documents were removed, ensuring their anonymity, they were uploaded to the Bismart Folksonomy portal, proceeding to their conversion to text using an OCR system from Microsoft Cognitive Systems, and using pattern detection algorithms, the different fields based on the sections of the reports (diagnoses, reason for consultation, personal history, usual treatment, complementary examinations, evolution and treatment at discharge), to later store the already anonymized information in the Hospital del Mar data cloud. Next, data normalization and lemmatization processes were carried out, algorithms were applied, folksonomization was performed, and the Bismart Folksonomy web portal was installed. Bismart has an automatic process that installs in the chosen data cloud a virtual machine with the database and all the necessary services for the Bismart Folksonomy portal; so that the technical deployment is a relatively simple process. (Fig. 1). The portal complies with all GDPR regulations (general data protection regulation) with a registry of accesses and modifications or inquiries made to the data; the time, the user and the IP address is recorded in a log of the device.

Fig. 1.

The solution proposed by Bismart is based on a flow of data that begins with the incorporation into the database knowledge of the PDF documents provided by Hospital del Mar. On these documents we apply OCR processes and the definition of the fields that we want to import from each document. Once the fields are stored in the database and the fields have been identified, the system starts the folksonomy process, detecting important words or groups of words in the collection of documents. Once the Folksonomy tool has extracted the information that is needed to work, it is presented in a Web so that it can be consulted, modified or to execute the process again upon request.

(0.16MB).

Medical terms and acronyms specific to the specialty appeared in the documents. Furthermore, the reports were written indistinctly in two languages (Catalan and Spanish) and terms in both languages could even appear in the same report. This added more complexity to the data extraction, but since folksonomy works with terms and not with languages, the creation of synonyms of words between both languages or of words and acronyms allows the identification of the search term regardless of the language.

In the field of Nephrology, the classification of the degree of chronic kidney disease (CKD) is very important5 since it has prognostic and therapeutic implications. The review of the classification of the disease or renal situation according to the information collected in the “diagnoses” section of the discharge reports, only allowed us to identify the degree of CKD in some 300 reports. Due to the fact that the tool allows the addition of synonyms, the words “grau”, “estádio” and “estadi” were assigned to the word “grado”, which made it possible to find the degree of CKD in 768 reports. To classify the degree of CKD in the rest of the reports, as well as to identify reports with the diagnosis of acute renal failure, algorithms with heuristic rules were generated for the correct identification of the renal disease situation based on: a) the presence of the words “acute renal failure” and synonyms in the diagnosis section, which implied the label of acute renal failure, b) the presence of the words “admission for kidney transplant recipient” and synonyms in the reason for consultation implied the label CKD grade 5, c) the identification of the words “chronic kidney disease grade X” and synonyms among the personal history allowed the reports to be labeled as CKD grades 1–5, d) the use of creatinine in the admission analysis with age and gender (data collected between the anthropometric variables) allowed the calculation of the estimated glomerular filtration rate by entering the CKD-EPI formula6 in the software (chronic kidney disease epidemiology collaboration). Despite this, the algorithms did not allow the classification of 79 documents in terms of renal status, so the renal status was manually reviewed and assigned in the remaining patients that were not classified. Thus, the application of the software tools (creation of labels, synonyms and algorithms with heuristic rules) allowed the automatic classification of the degree of renal disease despite not being included in the “diagnoses” section of the medical records in 95% of all reports, manual review was required only in 5%. In this way, all reports were classified as: acute renal failure, CKD grades 1–5 or without renal disease (Fig. 2).

Fig. 2.

Algorithm applied to classify the kidney disease status of the reports included in the analysis.

(0.32MB).

Despite the fact that metformin continues to be the most widely used oral hypoglycemic agent with the greatest evidence of efficacy in the treatment of type 2 diabetes due to its benefits in terms of morbidity and mortality, its use in patients with kidney disease is restricted.7 In CKD, the dose must be adjusted and its use is contraindicated in advanced CKD, as it may be associated with the presence of lactic acidosis, especially when administered to patients with a glomerular filtration rate less than 30mL/min/1.73m2.8 Thus, in a situation of advanced CKD, only other types of oral hypoglycemic agents can be used or it may be considered the use of insuline. The treatment received by diabetic patients admitted to the Hospital del Mar Nephrology service and the characteristics of patients diagnosed with metformin-induced lactic acidosis were analyzed.

Diabetic patients were identified based on the presence of this diagnosis in the “diagnoses” section of the discharge reports. Thus, it was identified a percentage of reports with a diagnosis of diabetes lower than expected, for which reason the search was expanded with new heuristic rules, assigning the diagnosis of diabetes to those reports in which some hypoglycemic drug was found in the treatment. Given the large number of hypoglycemic agents available on the market, the inclusion of each of them individually in the searches generated greater complexity for the project. Thus, to label each drug, it was decided to use the ATC (Anatomical Therapeutic Chemical classification system) classification of the Spanish Medicines Agency (AEMPS), which groups these compounds by the active ingredient. The ATC classification provides information on both the trade name and the active ingredient, and in the data analysis process the entire text was searched for either one. In the case of identifying a trade name, thanks to the ATC classification, the active principle and the ATC group to which it corresponds can be inferred by applying graph analysis algorithms. It must be taken into account that there are ATC groups that can contain two or more active ingredients, so this logic was added to the detection algorithm.

In the case of drugs for the treatment of diabetes, these correspond to group A, subgroup A10 of the ATC classification.

Finally, there were identified 637 of the 1631 reports with the diagnosis of diabetes (39.05% of the reports). Table 1 shows the hypoglycemic treatments received by these patients according to their degree of renal disease (subgroups were A10A for insulin, A10BA for metformin, A10BH for DPP4 inhibitors, A10BX for repaglinide and transporter the inhibitor SGLT2 –requiring search by trade name and active principle-, A10BB for derivatives of sulfonylureas and A10BG for thiazolidinediones). Thus, the most widely used treatment in these renal patients with diabetes is insulin (337 reports contained insulin in the usual treatment on admission), followed by metformin (85 reports contained metformin in the medication on admission). In five of these 85 reports, the term “lactic acid” and its synonyms were identified in the diagnostic section, thus revealing five cases of metformin-induced lactic acidosis (three episodes in the context of acute renal failure, one patient with stage 3 CKD, and one patient with stage 4 CKD). In all cases, the reason for consultation turned out to be acute gastroenteritis (four with diarrhea and one with emetic syndrome). In addition, in all cases, except for the patient with stage 3 CKD, hemodialysis was required in the context of impaired renal function and lactic acidosis due to metformin.

In 102 reports, it was not detected any hypoglycemic treatment, so it was concluded that 16% of the patients followed only dietary treatment for their diabetes.

The prevalence of arterial hypertension in kidney disease is high; both pathologies coexisting in 80%–85% of patients with kidney disease.9

The renal and cardiovascular benefits of renin angiotensin system inhibitors (RAS inhibitors) in patients with CKD have been widely demonstrated10 and they should be part of the treatment of arterial hypertension in renal patients. However, in situations of acutly decompensated renal function, they are usually withdrawn. The delay in their reintroduction once the decompensating episode has resolved could imply a worsening of the prognosis of our patients.11

It was analyzed the antihypertensive treatment received by patients diagnosed with arterial hypertension and admitted to the Hospital del Mar Nephrology Service. The diagnosis of arterial hypertension was identified in 1520 of the 1631 available reports (93.19% of the reports). For this, the synonyms “arterial hypertension” and “HTN” were included under the label “arterial hypertension”. The antihypertensive drugs that patients received separated by groups (atc C02A were centrally acting antihypertensives, atc C02C for doxazosin, atc C02DB for hydralazine, atcC03A for diuretics: 03AA hydrochlorothiazide, 03BA chlorthalidone and indapamide, 03CA loop diuretics and 03DA for antialdosterones; atc C07 for beta-blockers, atc C08CA dihydropyridine calcium antagonists and atc C08D non-dihydropyridines, and atc C09 for renin angiotensin system inhibitors). On admission, diuretics were the most commonly used drugs (in 754 reports there was at least one diuretic in the usual medication, mostly [549] loop diuretics). It should be noted that 34 of the 126 reports that recorded the administration of thiazide and related diuretics corresponded to patients with stage 5 CKD, a degree of kidney disease in which the diuretic effect of these drugs is lost. The second most widely used antihypertensives were the family of beta-blockers (611 reports), followed by dihydropyridine calcium antagonists (532 reports) and then renin-angiotensin system inhibitors (437 reports). Admission modified the pattern of antihypertensive administration at discharge, both in terms of the total number of antihypertensive drugs prescribed (2588 at admission and 2758 at discharge) and in the increase or reduction of prescription of certain families of antihypertensives, as shown in Table 2. In 437 of the 1520 reports with a diagnosis of arterial hypertension (28.75%), a drug belonging to the C09 group according to the ATC classification (ARS inhibitors) was identified in the usual treatment. The renal situation of the patients receiving treatment with this group of drugs on admission is shown in Table 3 (column 2). The same table (column 3) shows the number of reports that continued to receive RAS inhibitors at discharge, while the last column of the table shows the percentage reduction in the prescription of renin-angiotensin system inhibitors at the time of discharge from Nephrology. A withdrawal of RAS inhibitors stood out in a high percentage of discharge reports, showing an increasing trend depending on the degree of chronic kidney disease (from stages 3–5). In this study, the causes of withdrawal of these drugs were not analyzed, although knowing the usual clinical practice, it was probably related to the acute deterioration of renal function, more likely to observe with more advanced the renal disease. Taking this reasoning into account, it did not seem plausible with routine clinical practice that the percentage of reports maintaining treatment with RAS inhibitors at discharge among those classified as “acute renal failure” would be so high (only 25% reduction in prescription at discharge). For this reason, these reports were manually reviewed. The manual review made it possible to detect words such as “suspend” or “modify” in front of RAS inhibitor drugs, so that, in the acute renal failure group, only seven reports actually maintained the treatment at discharge, having therefore been erroneously detected and as false positives 11 reports.

There are studies showing that the prevalence of depression symptoms in patients with CKD is high and that psychosocial variables play an important role in the perception of quality of life in kidney patients.12,13 However, in the the clinical care of nephrology departments, psychological cures of our patients continue to be relegated to a secondary priority.

A search was made for those reports that contained in the usual treatment at admission a drug from group N05 or N06 according to the classification (psycholeptic and psychoanaleptic drugs). There were identified 402 reports including some of these drugs (24.6% of the reports). However, if the search was based on the presence of a diagnosis of the anxious-depressive field in the sections of “diagnoses” or “personal history”, only 45 (2.75%) and 192 (11.77%) were identified respetively. These data show that the physician's awareness regarding the prevalence of anxiety-depressive disorders in renal patients is low despite a high prescription of drugs to treat their symptoms.