DETECTION OF INCONSISTENCIES IN HOSPITAL DATA

CODING

Juliano Gaspar, Fernando Lopes and Alberto Freitas

CIDES – Department of Health Information and Decision Sciences, Porto, Portugal

CINTESIS – Center for Research in Health Technologies and Information Systems, Porto, Portugal

Faculty of Medicine, University of Porto, Porto, Portugal

Keywords: Inconsistencies, Errors, Data quality problems, Hospital databases.

Abstract: Introduction: Health professionals need data, in sufficient quantity and quality, and tools that can manage

the vast amount of available data. They need help for data management and appropriate support for decision

making.

Aim: The focus of this study is to develop a prototype that can contribute to the identification of data quality

problems in clinical and administrative data.

Methods: Methods involve the definition of requisites and business rules, the prototype development and

testing, and the realization of two studies using the prototype.

Results: Studies performed using the prototype resulted in the detection of many data problems and

inconsistencies. Amongst those we can point out, for instance, that 82,000 (15%) episodes had ‘diagnostic

code does not exist in ICD-9-CM table’ and that 783 (0,2%) episodes within ‘female breast cancer’ had the

variable gender equal to ‘male’.

Discussion: This prototype, besides contributing to the detection of data quality problems, is also expected

to be an incentive to the improvement of information system architectures. It shows the importance of the

development of mechanisms to detect and validate data in health environments.

1 INTRODUCTION

Medical databases generally involve all sorts of data

and store a considerable size of information (Kumar

et al., 2008). Data consists, generally, of records

with different type of information and with diverse

characteristics such as age, blood type, weight,

clinical images, diagnosis, lab results, among many

other patient details (Chandola et al., 2009). In the

latest years, with an exponential growth in hospital

information, the volume of electronic clinical

records has also significantly increased. Impaired

with this increase, new interests emerged in the

analysis of such information, thus driving it not only

to be used as a source to make decisions but also to

support hospital management (Freitas et al., 2005).

Amongst the various health services, hospital

focus gains a particular interest because it gathers a

higher complexity and procedures cost. This makes

hospital databases the main focus of various analysis

(Freitas et al., 2005). The display of reliable and

trustworthy information, obtained from solid data, is

vital to help health professionals and technicians in

the process of decision making and also help in

higher levels of hospital management. In this

context, Information Systems are responsible for

making, analyzing and disseminating such data

(Pinto, 2010).

1.1 Data Quality

The data value is directly proportional to its quality.

So, the higher the quality is, the greaters its utility

(Arts et al., 2002). It is important to consider that, to

insure data quality, certain basic rules must be

followed such as coherency, accuracy, integrity and

consistency (Tayi and Ballou, 1998

); (Wang, 1998);

(Silva-Costa et al., 2010). Thus, the importance of

rectify integrity problems, normalize values, fill or

identify missing values, identify redundant

information, among other processes in database

management (Barateiro and Galhardas, 2005).

The quality of data, accordingly to some authors,

is a relative concept and is directly correlated to the

189

Gaspar J., Lopes F. and Freitas A..

DETECTION OF INCONSISTENCIES IN HOSPITAL DATA CODING.

DOI: 10.5220/0003757301890194

In Proceedings of the International Conference on Health Informatics (HEALTHINF-2012), pages 189-194

ISBN: 978-989-8425-88-1

 2012 SCITEPRESS (Science and Technology Publications, Lda.)

information purpose. The authors use the “fitness for

use” concept that emphasizes the importance of

users’ point of view and their judgment in

information assertiveness (Tayi and Ballou, 1998

);

(Wang, 1998

); (Olson, 2003). Especially in medical

databases, this concept reinforces itself, because data

can have sufficient quality for economic analysis but

insufficient to clinical or epidemiological studies

(Silva-Costa, 2010

); (Cruz-Correia et al., 2009).

In the scenario of hospital department, electronic

data quality problems (DQP) gains a higher attention

because the information has a vital importance for

patient care (Daniel et al., 2008). However, due to a

high level of complexity and integrity inherent to

this activity, maintaining a high quality database is

not an easy chore.

1.2 Medical Data Coding Guidelines

There’s many disparity in health policies and

hospital funding between different countries.

Despite those differences, there are common

methods, characteristics and needs in all hospital

environments which make it possible to define a set

of international rules to analyze and compare

different hospitals in different countries, having in

consideration its’ health records (Pinto, 2010).

After being collected and coded, medical data

concerning patients, usually has billing information

frequently used only for administrative and financial

purposes. This does not prevent them from being

usefull to other areas such as health quality and

health care (Price et al., 2003).

These databases have a huge statistical potential,

due to their large volume and similarity between

hospitals, intitutions or even countries. This

comparation between hospitals is due, mainly, to the

use of a core group of variables named

internationaly as “Minimum Basic Data Set”

(MBDS) (Aylin et al., 2007

); (Romano et al., 1995).

A part of this coded data is generated in the

hospital discharge codification process. In this

process, the information collected from clinical

records, mainly diagnosis and procedures, is

generally coded according to the “International

Classification of Diseases” (ICD). It is also recorded

the “Diagnostic Related Group” (DRG) in which

episode is grouped. The use of ICD-9-CM

, along

these past years, has had many purposes such as



ICD-9-CM: International Classification of Diseases, Ninth

revision, Clinical Modification.

supporting the health services payment, studding

their costs, evaluate quality, planning future needs

and helpping clinical research (Ginde et al., 2008),

among others.

In Portugal, systematic codification of

hospitalizations in 1989 (Lopes, 2010) and it was

addopted the use of DRG as a hospital “case mix”

measurement. Nowadays, the main hospital

information systems have databases made of clinical

and administrative data, which were compiled

accordingly to the European established MBDS

(Oliván, 1997).

In Portugal, the Health Ministery defines and

publishes regulations that, not only, determine the

prices to be charged by the health delivered in the

National Health Service (NHS), but also clearly

distinguishes between what should be an inpatient or

an ambulatory episodes, among other rules for the

codification in DRG.

The evaluation of clinical coding quality is

essential for the correct attribution of hospital

funding, as hospital evaluation and funding

theirselves depend in DRG grouping (Pinto, 2010).

So, all MBDS potential for hospital management

and research can be compromised if clinical

codification does not have certain levels of quality.

2 AIM

The aim of this study is to develop a prototype for

the detection of data quality problems and

inconsistencies in inpatient and ambulatory episodes

in hospital databases. The prototype is intended to

detect common errors in health related databases and

also contextualize errors in specific medical

specialties.

3 METHODS

Prototype and the studies regarding its use the

development process follow the following

predefined steps:

 Analyze the central database structure and the

coding guidelines;

 Identify the business rules, define prototype

requisites, implement and realize the tests;

 Perform a two studies including data from the

last 10 years, the first from a hospital and second

including data concerning a pre-determined disease

diagnosed in all NHS hospitals.

HEALTHINF 2012 - International Conference on Health Informatics

190

The database used is composed of hospital episodes,

data ambulatory and inpatient data, from medical or

surgical specialties, concerning 96 Portuguese

hospitals with discharges between the years 2000

and 2009, making a total of approximately 12

million records. Access to this data was possible due

to the ACSS

collaboration in the HR-QoD

project

developed by the Department of Health Information

and Decision Sciences of Faculty of Medicine of

University of Porto. All data used was anonymous

and clinical data was coded in ICD-9-CM.

Based on the defined requisites and business

rules, auxiliary tables were created to store

information from the DQP detected such as error

codes and the errors quantity detected in each

hospital and year. Business rules contain, rules that

validate whether the information contained in a

record is consistent, meaning that it does not present

incongruences between the values of a variable in

relation to values of other variables in the same

episode. If the values are accordingly to regulations

and guidelines of that codifications used in Portugal,

such as ICD-9-CM. These were created and

schematized as showed in figure 1.

Figure 1: Business rule to variable “AGE”.

In the implementation of the prototype the

following programming languages were used: PHP,

HTML and JavaScript the Eclipse developing tools

for PHP and Toad for Oracle were also used.



ACSS: Administração Central do Sistema de Saúde (in

portuguese).

HR-QoD: Quality of data (outliers, inconsistencies and errors) in

hospital inpatient databases: methods and implications for data

modeling, cleansing and analysis.

4 RESULTS

With its’ use it was possible to create reports with

the detection of DQP and respective incongruences.

These reports included the analysis of over 50

different types DQP in the studied data.

As result of a first study, data from a hospital

concerning discharges between the years 2000 and

2009 (inclusive) was selected, totalizing 543,133

hospital episodes. Some of the problems detected are

shown in table 1.

Table 1: DQP Detected.

Problem Description N %

Incorrect length of stay (LOS) 191,282 35.22

Diagnostic code not in

ICD-9-CM

table 82,422 15.18

Procedure code not in

ICD-9-CM

table 9,023 1.66

Principal diagnostic is null 85 0.02

The evolution, in the 10 years of determined

detected DQP can be seen in graphics presented in

figure 2 and 3.

Figure 2: Data Quality Problems within 10 years.

In the figure 2 we can see that, for example, the

growth (after 2007) errors in the variable ‘AGE’

(calculated incorrectly) and also that the total days in

preoperative are incorrectly calculated after 2004.

The figure 3 show a gradual decline of errors

related to ‘weight at birth’, the accentuated increase

of ‘repeated diagnosis’ in the same episodes between

2007 and 2008, the gradual increase related to ‘inter-

hospital transfers’ and identical volumes of errors

between 2000 and 2004 related with ‘LOS sum

(preoperative)’ and ‘surgical date’.

DETECTION OF INCONSISTENCIES IN HOSPITAL DATA CODING

191

Figure 3: Data Quality Problems within 10 years.

For the second study, the pathology Female

Breast Cancer (FBC) was chosen due to a publication

released in January 2011 by INE

, stating that in

Portugal 1,635 people died in the year 2009 (INE,

2011) due to FBC and according to WHO

, FMC is

the most frequent tumor worldwide. It’s the main

cause of female death internationally (Bastos et al.,

2007); (WHO, 2011).

To this purpose, all episodes with diagnosis of

FBC coded in ICD-9-CM with ‘174.x’ were selected

in all 96 Portuguese hospitals in the last 10 years,

resulting 346,654 records.

Specific queries were applied in order to detect

inconsistencies in the context of codification in this

disease. Table 2 shows the most relevant

inconsistencies detected.



INE: Instituto Nacional de Estatística (in Portuguese)

WHO: World Health Organization.

Table 2: Contextual DQP Detected.

Contextual Problem Description N %

‘V58.0’ in secundary diagnostics 66,875 19.34

‘V58.1’ in secundary diagnostics 72,281 20.90

‘V10.x’ in principal diagnostic 830 0.24

Episodes with ‘gender’ variable = ‘male’ 783 0.23

From this analysis we can point out 783 records

with value “male” in the variable ‘gender’. This fact

represents a serious DQP because this study is

specifically about FBC. Male Breast Cancer is coded

in ICD-9-CM with ‘175.x’, and so code ‘174.x’ is

restricted to female patients.

5 DISCUSSION

The algorithm was capable of identifying the most

significant DQP existing in the database.

The “Incorrect length of stay (LOS)”, one of the

DQP detected in 35% of total episodes, observes

some particular considerations: (a) the regulation

number 839-A/2009 from the health ministry

informs that “Global Length of Stay” (G-LOS)

concerns the total number of days consumed by each

patient within the hospital, considering the

admission day and ignoring the discharge day (SNS,

2009); (b) a most detailed analysis of this

inconsistency value showed that, despite the

business rule is correct with the in vigor regulation,

this value can be caused by inconsistencies in

informatics applications that calculate the variable

G-LOS; (c) some specialists make their calculations

only with the value of the day in question;

meanwhile other consider also the hours and

minutes, which can cause one day difference in G-

LOS.

DQP occurrences DQP “Diagnostic code not in

ICD-9-CM table” and “Procedure code not in ICD-

9-CM table”, showed in table 1, are due essentially

with the passing years, to the reviewing of some

codes and the modification or others. However, the

process associated to the management of these

modifications is not performed efficiently. This

process would have to store the different versions of

changes made to the ICD-9-CM table used in each

hospital as well as to store the version used in the

codification process instant (Lopes, 2010).

Codification errors on principal diagnosis, such

as the 85 cases presented on table 1, imply the

classification of episodes with DRG 470

(“Ungroupable”) in these cases, the hospital is not

financially compensated for the episode costs (Silva-

Costa et al., 2010).

HEALTHINF 2012 - International Conference on Health Informatics

192

The gradual decrease of DQP related with

‘weight at birth’, presented on figure 1, could be

associated with data validation rules, performed

during data entry, by the different electronic

applications used on NHS hospitals during the last

years.

In figure 3 a gradual increment related to patient

transfers between hospitals after the year 2006 can

be observed. DQP “Code of the reason for the inter-

hospital transfer is null” and “Code of the inter-

hospital transfers is null”, raise the issue that the

lack of these codes could be related to regulations

that the health ministry posted in 2003: (a) a hospital

that transfers a patient to other, by “lack of

resources”, can only receive up to 50% of the

correspondent DRG; (b) the hospital that receives a

transferred patient is limited, in its’ classification, to

a restrict set of DRG.

Despite that these regulations already existed in

2003, only after 2006 all NHS hospitals had these

regulations. Regardless of the cause, it is a fact that

this problem has obvious implications in hospital

reimbursement level.

An analysis to identify the existing relation

between errors found in years 2000 to 2004, in DQP

sum (preoperative) < 0’ and ‘Surgical date <

Admission date’ (Figure 3), revealed that all related

episodes had many patients that were operated in the

emergency service and then admitted in the hospital.

After 2004 the ‘Admission date’ to be considerate

was the lesser date registered, being the emergency

service entry date or the admittance date in the

hospital services.

The identification of 783 cases coded with

‘174.x’ and with the ‘gender’ variable equal to

‘male’, in the second study, is of extreme

importance. Assuming that these records are actually

male patients, this inconsistence represents a 0.23%

impact in this study. However, if the focus of the

studies was Male Breast Cancer, coded in ICD-9-

CM as ‘175.x’, these 783 cases wrongly coded

would correspond to a total of 25% cases of Male

Breast Cancer that may not be considered.

Official coding guidelines (ACSS, 2009) state

that “in admittance of the patient exclusively to

chemotherapy, imunohemoterapy or radiotherapy,

the principal diagnosis code attributed should be

‘V58.x’, and secondary diagnosis should be the ones

representing the cause for treatment”.

After these directives, we can verify that in

139,156 records, 40.2% of total episodes of FBC

(Table 2) present ‘V58.0’ or ‘V58.1x’ as secondary

diagnosis codes. The average rate of this

inconsistency is 64% per year. The directives also

state that the code ‘V10.x’ (‘personal history of

malign cancer’) should always be coded as a

secondary diagnosis, however 830 records were

found with ‘V10.x’ was the principal diagnosis.

6 CONCLUSIONS

This study helped the analysis and identification of

existing DQP in a coded health database, and also

helped in the determination of a profile for the

coding of FBC episodes.

Acknowledge of anomalies in the data may allow

an immediate improvement in their quality, when

the correction is feasible. This acknowledgement is

important for the management of health, both

nationally and regionally, and also locally in each

hospital or health facility.

The results are graphically simple and allow a

direct reading of the most relevant values and are

certainly an asset to help managing and controlling

the efficiency of health facilities.

The visualization of DQP, in graphical format

over 10 years, allowed users, only with a visual

analysis of the evolution of the problem, to quickly

identify possible relationships between the increase

or decrease of a problem and the date of

implementation of certain software in the hospital,

or the adoption of a new standard for procedure..

Effective management of versions, also known

as change management, applied both by managers

and by the NHS hospital managers, in addition to

companies that develop software for health, can help

to prevent problems to happen again, as some of

those presented in this study.

This study is a contribute to the detection of

DQP, and also an incentive to the improvement of

the architecture of existing information systems,

emphasizing the importance of developing

mechanisms and methods for detection and

validation in health. This study also intends to alert

for the impact of these errors in epidemiologic

studies and, as well, in management evaluations and

health related policies.

6.1 Future Works

In the next stages of this study, we intend to

replicate the method used in this study to other

pathologies and apply data mining algorithms to

improve the detection of patterns, abnormal cases,

possible errors or inconsistencies.

DETECTION OF INCONSISTENCIES IN HOSPITAL DATA CODING

193

ACKNOWLEDGEMENTS

The authors would like to thank the support given by

the research project HR-QoD - Quality of data

(outliers, inconsistencies and errors) in hospital

inpatient databases: methods and implications for

data modeling, cleansing and analysis (project

PTDC/SAU-ESA/75660/2006).

REFERENCES

ACSS, (2009). CID-9-MC Guidelines Oficiais para

Codificação: Em vigor a partir de Outubro de 2009

(Lisboa, Administração Central do Sistema de Saúde,

IP).

Arts, D., Keizer, N., and Scheffer, G.-J., (2002). Defining

and Improving Data Quality in Medical Registries: A

Literature Review Case Study, and Generic

Framework. In: J Am Med Inform Assoc 9, 600-611.

Aylin, P., Bottle, A., and Majeed, A., (2007). Use of

administrative data or clinical databases as predictors

of risk of death in hospital: comparison of models. In:

BMJ 334, 1044.

Barateiro, J., and Galhardas, H., (2005). A Survey of Data

Quality Tools. In: Datenbank-Spektrum 5, 15-21.

Bastos, J., Barros, H., and Lunet, N., (2007). Evolução da

Mortalidade por Cancro da Mama em Portugal (1955-

2002). In: Acta Med Port, 139-144.

Chandola, V., Banerjee, A., and Kumar, V., (2009).

Anomaly Detection: A Survey. In: ACM Computing

Surveys 41.

Cruz-Correia, R., Rodrigues, P., Freitas, A., Almeida, F.,

Chen, R., and Costa-Pereira, A., (2009). Data Quality

and Integration Issues in Electronic Health Records. In

Information Discovery on Electronic Health Records.

In: CRC Data Mining and Knowledge Discovery

Series, H. V. C.a. Hall, ed., pp. 55-95.

Daniel, F., Casati, F., Palpanas, T., Chayka, O., and

Cappiello, C., (2008). Enabling Better Decisions

through Quality-aware Reports. In: International

Conference on Information Quality (ICIQ) (USA).

Freitas, A., Brazdil, P., and Costa-Pereira, A., (2005).

Mining Hospital Databases for Management Support.

Paper presented at: IADIS Virtual Multi Conference on

Computer Science and Information Systems, pp. 207-

212.

Ginde, A. A., Tsai, C. L., Blanc, P. G., and Camargo, C.

A., Jr., (2008). Positive predictive value of ICD-9-CM

codes to detect acute exacerbation of COPD in the

emergency department. In: Jt Comm J Qual Patient

Saf 34, 678-680.

INE, (2011). Boletim mensal de estatística: Janeiro de

2011. Available in: http://www.ine.pt. Access In:

02/03/2011.

Kumar, V., Kumar, D., and Singh, R. K., (2008). Outlier

Mining in Medical Databases: An Application of Data

Mining in Health Care Management to Detect

Abnormal Values Presented In Medical Databases. In:

IJCSNS International Journal of Computer Science

and Network Security, 272-277.

Lopes, F., (2010). Portal da Codificação e dos GDH.

Available in: http://portalcodgdh.min-saude.pt.

Access In: 31/10/2010.

Oliván, J. A. S., (1997). Sistemas de información

hospitalarios: el CMBD. In: Scire: representación y

organización del conocimiento 3, 115-130.

Olson, J. E., (2003). Data Quality - The Accuracy

Dimension, Morgan Kaufmann Publishers edn.

Pinto, R., (2010). Sistemas de informações hospitalares de

Brasil, Espanha e Portugal - Semelhanças e diferenças.

In: FIOCRUZ (Rio de Janeiro, Escola Nacional de

Saúde Pública Sergio Arouca), pp. 162.

Price, J., Estrada, C. A., and Thompson, D., (2003).

Administrative Data Versus Corrected Administrative

Data. In: Am J Med Qual 19, 38-44.

Romano, P. S., Zach, A., Luft, H. S., Rainwater, J., Remy,

L. L., and Campa, D., (1995). The California Hospital

Outcomes Project: using administrative data to

compare hospital performance. In: Jt Comm J Qual

Improv 21, 668-682.

Silva-Costa, T., (2010). Indicadores de Produção

Hospitalar - Uma forma de medir a produção dos

hospitais Portugueses (Porto, Faculdade de Medicina

da Universidade do Porto), pp. 172.

Silva-Costa, T., Marques, B., and Freitas, A., (2010).

Problemas de Qualidade de Dados em Bases de Dados

de Internamentos Hospitalares. Paper presented at: 5ª

Conferência Ibérica de Sistemas e Tecnologias de

Informação (Santiago de Compostela).

SNS, (2009). Portaria n.º 839-A/2009 - Ministério da

Saúde de Portugal, M.d. Saúde, ed. (Lisboa, Diário da

República), pp. 4978-(4972) a 4978-(4124).

Tayi, G. K., and Ballou, D. P., (1998). Examining Data

Quality. In: CACM 41, 54-57.

Wang, R. Y., (1998). A Product Perspective on Total Data

Quality Management. In: CACM 41, 58-65.

WHO, (2011). World Health Organization. Available in:

http://www.who.int/en/. Access In: 20/07/2011.

HEALTHINF 2012 - International Conference on Health Informatics

194