Expanding medical knowledge increases the potential risk of medical errors in clinical practice. We present, OPAD, a clinical decision support system in the field of the medical care of osteoporosis. We utilize clinical information from international guidelines and experts in the field of osteoporosis. Physicians are provided with user interface to insert standard patient data, from which OPAD provides instant diagnostic comments, 10-year risk of fragility fracture, treatment options for the given case, and when to offer a follow-up DXA-evaluation. Thus, the medical decision making is standardized according to the best expert knowledge at any given time. OPAD was evaluated in a set of 308 randomly selected individuals. OPAD’s ten-year fracture risk computation is nearly identical to FRAX (
According to the International Osteoporosis Foundation (IOF), one in three women and one in five men will experience an osteoporotic fracture later in their life; thus, globally an osteoporotic fracture is estimated to occur every third second [
The diagnosis of osteoporosis has traditionally relied on bone mineral density (BMD) measurement [
Currently available risk calculator systems for osteoporosis, for example, FRAX [
We have developed a clinical decision support system that gives the 10-year fracture probability due to osteoporosis for the individual patient, lifestyle recommendations, and recommendations as to whether and when a BMD scan is recommended. Furthermore, the system will identify patients at risk of fractures and would benefit from specific preventive medical treatment. Our osteoporosis adviser (OPAD) is designed for those with medical knowledge such as general physicians and clinical nurse specialists.
In this paper, we evaluate the reliability of the OPAD system by comparing its 10-year fracture probability with the probability given by FRAX and by assessing the quality of its BMD scanning recommendations, that is, whether those that were recommended to have a BMD measurement benefited from the measurement.
We have designed an expert system to assist in the diagnosis and treatment of osteoporosis. The software takes as input a set of clinically relevant parameters from which the 10-year fracture risk is computed based on published country specific data [
The software further outputs immediately relevant information to the user: a risk group for the individual (low, medium, or high risk for fracture compared to age-matched controls [
The design of the system uses a knowledge mapping approach. Expert physicians were queried to determine the clinically relevant parameters for the recommendation of osteoporosis treatment and recommendations for BMD measurements. A group of different specialists who were all interested in osteoporosis (rheumatologist, endocrinologist, general practitioner, and geriatrician) participated in the process. The Intellix Advisor [
The Intellix Advisor allows for active acquisition of knowledge. A set of instances is input into the software and from these input examples the software can be told either to construct a neural network based model based on the examples or to ask for more examples that consist of patients not considered by the model. In our approach patients were added to the model until a diagnosis or recommendations could be made for every possible tested patient. The model is implemented as a lookup table and for every new patient diagnosed a patient with the same characteristics is found in our database.
The diagnosis for a patient includes four distinct pieces of information: 10-year fracture risk, lifestyle recommendations, treatment recommendations, and recommendation of the time for the next BMD measurement or follow-up evaluation of the individual patient. Initially a set of clinically relevant parameters was determined (see Table
Patient attributes used by the osteoporosis advisor.
Age |
Bone mineral density ( |
Ethnicity |
Gender |
Previous osteoporotic related fracture |
Parent hip fracture |
Current smoking |
Current use of glucocorticosteroids for more than three months |
Rheumatoid arthritis |
Secondary osteoporosis |
Alcohol: 3 or more units per day |
Hormone replacement therapy |
Regular exercise |
Sufficient calcium intake |
Sufficient vitamin D intake |
The OPAD system follows the frame of international guidelines, for example, the Scottish Intercollegiate Guidelines Network (SIGN) on osteoporosis (
Table
Patient attributes used for the recommendation of osteoporosis treatment.
Attribute | Type |
---|---|
Gender | Male/female |
GIOP | Yes/no |
Fragility fracture | Yes/no |
Fracture risk | High/medium/low |
Treatment | Yes/no |
Secondary osteoporosis | Yes/no |
|
Numerical |
Age | Numerical |
Menopause status | Before/<3 years/>3 years |
Diagnosis | None/osteopenia/osteoporosis/manifest osteoporosis/GIOP |
The recommended time until the next DXA measurement was also determined by using a knowledge mapping process. A list of clinically relevant pieces of information was determined which can be seen in Table
Attributes used to determine time until next DXA measurement.
Attribute | Type |
---|---|
Gender | Male/female |
Menopause | Before, <3 years, >3 years |
Risk group | High/medium/low |
Treatment | Yes/no |
Changes in BMD measured by DXA | No DXA/improving/unknown or losing/fast loosing/neutral |
Glucocorticosteroids | Yes/no |
Seven different recommendations were made for the time for the next BMD measurement, listed in Table
The possible recommendations for the next time for a BMD scan.
At menopause |
At the age of 65 |
Now |
In 1-2 years |
In 3 years |
In 5 years |
DXA not recommended |
The computed 10-year risk of fracture was based on the World Health Organization (WHO) fracture risk assessment recommendations [
In order to validate that the treatment recommendations presented to the end user agreed with the treatment recommendations originally determined for each patient a quality control module was developed. Thus, built on top of our clinical database a set of 300 virtual quality control patients was created which were used to automatically verify the correctness of the system recommendations for real life clinical information given for each case. Experiments verified that the results of these patients agreed in both the model created and the interface to the end user.
As the WHO recommendation guidelines are not given with a closed form formula we compared the fracture risk computation given by our OPAD system with the fracture risk computation given by FRAX. We selected consecutive 308 individuals from the out-patient osteoporosis clinic at LSH, who visited the clinic from the 1st of January 2012 onwards. We compared the ten-year fracture risk computed using the OPAD and the ten-year fracture risk computed using recommendations given by our model with those given by FRAX. Linear regression was run to compare the results between the two systems.
We also reevaluated the same group of 308 individuals with our OPAD system with respect to the need of DXA at the present time or later; that is, the risk evaluation was done without the DXA results (
The Data Protection Authority (S5680) and the National Bioethics Committee of Iceland approved the study protocol (VSNb2010050008). The original patients’ data were hosted by the University Hospital, Reykjavik, Iceland. The data were anonymized before being provided to the researchers.
Of the 308 cases, 39 were males and 269 were females, with a mean age of 61 years (15–89). Figure
Ten-year risk for major osteoporotic fracture computed using the osteoporosis advisor (OPAD;
These 308 patients were reevaluated by the OPAD with respect to the need for a DXA at the present time or later, that is, before they underwent their DXA evaluation.
In 178 cases (58%), out of these 308 cases, the OPAD system recommended DXA evaluation at the present time. Following DXA measurement in these 178 cases, 91 (51%) of those received OPAD recommendation on specific treatment options, where 5 patients (3%) were recommended to continue with their treatment. Additional 31 patients (17%) received recommendation on consulting specialist in osteoporosis. Meanwhile, only 51 of these 178 cases (29%) received general prevention measurement recommendations. Thus, the DXA investigation performed according to the recommendation of the OPAD system seems to influence the clinical decision-making process.
Out of the 308 original cases, 102 cases (33%) came for their DXA even though the OPAD system would have recommended that they should have their DXA at the age of 65; that is, every third patient who came in for a measurement did not need the DXA evaluation at the present time according to the OPAD system.
In only six of these 102 cases (5.9%), did the OPAD system change its recommendation following the BMD measurement? In four cases the OPAD recommended specific bone protective treatment, due to the fact that these four individuals were diagnosed with osteoporosis with a significantly increased risk of fragility fracture, that is, with a 10-year fracture risk in the range from 9.1% to 14.3%. In two additional cases the OPAD changed its recommendation to continuation of already taken measurements.
In a further 22 cases of these 308 cases (7.1%) the OPAD system recommended DXA within 1–3 years depending on various clinical circumstances. Independent of these recommendations all patients received a DXA measurement, like other patients in this study and only one of these individuals received a different recommendation following the DXA evaluation.
Osteoporosis is a disorder affecting the density and infrastructure of the bone mass [
Bone mineral density (BMD), that is, bone mass, may be measured with several methods, for example, quantitative ultrasound (QUS), peripheral quantitative computer tomography (pQCT), and dual energy X-ray absorptiometry (DXA) which is the golden standard of bone mass measurement. Most professional interest groups, for example, NOF and IOF, have published recommendations on when to use DXA for BMD measurement [
Several risk calculator tools for bone fractures have been presented, but only FRAX has been recommended and supported by the World Health Organization (WHO). FRAX was developed for the calculation of the 10-year fracture risk for hip fracture or a major osteoporotic fracture (clinical spine, forearm, hip, or shoulder fracture) based on certain risk factors, with or without results of DXA measurement of the hip. FRAX offers country specific values for several countries in Europe, North and Latin America, the Middle East, Africa, and Oceania, or a total of 52 country specific datasets [
The busy clinician may have difficulties in interpreting the risk value figure for each patient in hectic daily clinical practice. With this in mind we have extended the information provided by OPAD, by giving a specific diagnosis, that is, osteoporosis or osteopenia, and specific recommendations on prevention, time of next DXA, and treatment options according to international guidelines and experts knowledge [
A 62-year-old Swedish female patient with history of fracture whose mother also had a history of hip fracture. She did not have other medical risk factors and she had osteopenia according to a recent DXA evaluation with a
Although physicians subscribe to several medical journals, presenting thousands of articles yearly, they have difficulty in keeping up to date in all areas in their daily practice. The more complicated medical world amplifies the risk of diagnosis errors. OPAD allows “best practice” in osteoporosis risk evaluation of fragility fractures and treatment to be captured, distributed, and automated in a simple bedside manner for the busy practicing physician and other health care providers, including nurses working in fracture liaison services, as now highly recommended by IOF [
We conclude that OPAD is accurate in respect to fracture risk probability evaluation and may presumptively be cost effective in fracture liaison services. However, cost-benefit studies are needed in the field of osteoporosis preventive care and CDSS.
The authors declare that there is no conflict of interests regarding the publication of this paper.
Aron Hjalti Bjornsson, Bjorn Gudbjornsson, Bjorn Runar Ludviksson, Elvar Orn Birgisson, and Haukur Tyr Gudmundsson collected the patient data used in the paper. Aron Hjalti Bjornsson, Bjarni V. Halldorsson, and Haukur Tyr Gudmundsson analyzed the data collected. Bjarni V. Halldorsson, Bjorn Runar Ludviksson, and Bjorn Gudbjornsson wrote the paper. All authors read and approved the final version of the paper.
The authors would like to thank Klaus Bjorn Jensen, Thorsteinn Geirsson, Dagrun Jonasdottir, Elias R. Ragnarsson, and Orn Bardur Jonsson for their contribution to this work.