Development of a predictive Antibiotic Burden Index for Primary Immunodeficiency – an explorative study

(1)

Development of a predictive Antibiotic Burden

Index for Primary Immunodeficiency – an

explorative study

Milan Al-Naqshbande

Degree Project in Pharmacotherapy, 30 hp, Autumn semester 2019

Supervisor: Joakim Söderberg Examiner: Karin Svensberg

Division for Pharmacotherapy

Department of Pharmaceutical Biosciences Faculty of Pharmacy

(2)

Table of contents ____________________________________________________________ 2 Abstract ____________________________________________________________________ 3 Popular science summary _____________________________________________________ 5 Introduction _________________________________________________________________ 6 Primary immunodeficiency _________________________________________________ 6 Common Variable Immunodeficiency ___________________________________ 6 X-linked agammaglobulinemia _________________________________________ 6 Treatment of immunodeficiency _____________________________________________ 7 National Quality Registry for Primary Immunodeficiency __________________________ 7 Challenges in modern treatment of immunodeficiency ___________________________ 8 National Prescription Database _____________________________________________ 9 The future of healthcare technology ________________________________________ 10 The aim of the study _____________________________________________________ 11 Methods ___________________________________________________________________ 11 Outcome measures _____________________________________________________ 11 The process of obtaining the data __________________________________________ 12 Developing the Antibiotic Burden Index ______________________________________ 12 Interpretation of data ____________________________________________________ 14 Results ____________________________________________________________________ 14 ABI based on number of collected antibiotics – model 1 _________________________ 14 ABI based on ranks of antibiotics – model 2 __________________________________ 17 Discussion ________________________________________________________________ 20 Conclusion ____________________________________________________________ 22 Acknowledgements _________________________________________________________ 22 References ________________________________________________________________ 23 Appendix __________________________________________________________________ 25 Appendix A ____________________________________________________________ 25

(3)

Abstract

Introduction: Primary immunodeficiency is a medical condition characterized by frequent infections as a result of the immune system not functioning properly. Patients are usually treated with antibiotics and immunotherapy. Since there is no mutual database for patient records between primary care and healthcare, a communication gap arises. The idea behind the project is to investigate if there is a possibility to build a warning system that can act as an indicator for healthcare if antibiotics are being prescribed too often for these patients by primary care. The aim of this study was to examine whether it is possible to develop a predictive Antibiotic Burden Index (ABI) based on data generated from primary care.

Methods: In the study two models were designed and tested to see if they can describe how patients’ antibiotic use is related to levels of IgG. The correlations were evaluated to see if they could be used to design a warning system that would fire a signal if the patient is using a lot of antibiotic because that would be an indication of their treatment not being effective. Each individual antibiotic was given a value and from that

combined with the patients’ actual prescription fulfillment an ABI could be calculated. Two models for calculating ABI were evaluated, the first model is based on number of prescriptions collected from the pharmacy. The second one based on number of prescriptions plus an antibiotic score calculated using an antibiotic ranking system provided by Huddinge Hospital. These calculations were made using historical patient data from the last two decades. The data was extracted from the National Quality Registry for primary immunodeficiency.

Results:The results show that both models were applicable. The two models differed slightly in the percentage values, but both follow the same pattern. In non-Stockholm regions, the antibiotic use was higher during the six months up to the lowest recorded IgG and lower during the same time up to the highest recorded IgG. When evaluated in Stockholm only, it was strangely the opposite.

Discussion: One reason among others for the deviating results of Stockholm’s region could be a change in treatment recommendation within the region. Since this is an explorative study and the results seemed promising enough, it is recommended that the

(4)

project is taken into the next step. That would be a more profound study including more variables.

Conclusion: There seems to be a correlation between the use of antibiotics and IgG-levels.

(5)

Popular science summary

Antibiotikaanvändning inom primär immunbristsjukdom – ett nytt

samband upptäcks

Inom vården har de stora och vanliga sjukdomarna tagit över uppmärksamheten. Hypertoni, diabetes och psykiska sjukdomar är de vanligaste som kommer på tal nuförtiden. En sjukdom som inte är lika vanlig men som är viktig att skina

uppmärksamhet på är primär immunbristsjukdom. Den kännetecknas av återkommande och långvariga infektioner, bland annat lunginflammation och öroninflammation. Dessa infektioner uppkommer på grund av att kroppens immunförsvar är nedsatt och inte kan skydda mot mikroorganismer.

Behandlingen består av två delar, ökning av immunförsvaret samt hanteringen av infektioner med antibiotika. Ökningen av immunförsvaret görs via injektioner av just de komponenter i immunförsvaret som du har brist på. En av dessa komponenter är s.k. immunglobuliner som är en typ av antikroppar. Förr i tiden fick man gå till sjukhuset och få en injektion varannan vecka, där sjukvården även passade på att följa nivåerna av behandlingen i kroppen. Nuförtiden, i och med teknikens utvecklig, sker behandlingen hemifrån av patienterna själva. Frågan är då, hur kan vården ta reda på om patientens behandling går bra? Det kan de inte. Tiden mellan uppföljningar är två år lång, och så mycket kan hända under dessa två år utan vårdens vetskap.

Under dessa två år, om patienten en får infektion, är det vårdcentralen om tar hand om det och skriver ut antibiotika. I min studie, tittar jag på hur antibiotikauttagsfrekvensen beror på nivån av immunglobuliner i kroppen. Resultat på patienter i Sverige visar att ju högre nivåer det finns i kroppen, desto mindre antibiotika får den patienten. Detta beror på att patienten då inte får infektioner lika ofta och därmed minskar behovet av

antibiotika.

Resultaten i min studie ser lovande ut, det verkar finnas ett samband! Självklart behöver fler och mer djupgående studier göras på detta, men om resultaten fortsätter vara

lovande, har vi eventuellt lyckats hittat en metod som sjukvården kan använda sig av för att följa patienterna. Vi skulle då kunna bygga ett varningssystem där sjukvården får en varningssignal att kalla in patienten när patientens antibiotikauttag börjat öka, då detta skulle vara en indikation på att patientens behandling kan behöva justeras.

(6)

Introduction

Primary immunodeficiency

Primary immunodeficiency disease (PID) is a medical condition where one or more components of the immune system do not function properly (1). Among the most common of these, is a lack of immunoglobulins of various types (2). This type of disease has no specific symptoms but is usually characterized by recurrent and frequent infections. Usually, the infections occur in the ears and lungs, but could also appear in other parts of the body. It is important to be able to diagnose PID as early as possible in order to avoid permanent damage on the frequently infected organs (3). PID is

genetically caused and often comes to light early in life. There are over 300 known genes where the mutation in these causes different kinds PID with varying severity and courses of disease (4). Common variable immunodeficiency (CVID) and X-linked agammaglobulinemia (XLA) are two examples of these. There are over 130 known PID disorders caused by these mutation and new disorders are continuously being

recognized (1).

Common Variable Immunodeficiency

One of the most clinically significant forms of PID is CVID (5). CVID is characterized by hypogammaglobulinemia caused by dysfunction in the B-lymphocytes. The

symptoms and severity of the immunodeficiency in CVID is variable, i.e. they vary and appear differently in each individual (6). The disease is treatable, but the treatment is lifelong and consists primarily of the supply of immunoglobulins by infusion. Measures such as antibiotic therapy may be needed as prophylaxis if infections often occurs (6). X-linked agammaglobulinemia

Another clinically significant form of PID is XLA, also called Bruton’s disease, where the body does not produce enough B-cells, thus resulting in a lack of immunoglobulins of the type IgG (gamma globulin) (7). XLA is an inherited disease with the defect located in the BTK gene on the X-chromosome. The BTK gene is responsible for producing BTK protein which is essential for creating B-cells and for the immune system to function well. Since this inherited in a X-linked recessive pattern, the disease is more common in males because males only inherit one X-chromosome. If the inherited X-chromosome contains a mutated BTK protein, the subject will develop XLA. Meanwhile in females, BTK genes in both X-chromosomes has to be defected for the disease to develop, which is much more rare that in males (7).

XLA occurs in approximately 1 in 200,000 newborns. During the first months the child will seem healthy, due to the antibodies acquired by the mother pre-birth. It’s when these antibodies are cleared from the child’s body that the recurrent infections start to occur (7). Frequent otitis, sinopulmonary, diarrhea and skin infections caused by Streptococcus pneumoniae and Haemophilus influenzae are the most common

indicators of XLA (8). Although XLA-patients are sensitive to bacterial infections, this does not particularly mean that they are vulnerable to viral infections (7).

(7)

Treatment of immunodeficiency

Treatment of PID focuses on two aspects, i.e. managing infections and boosting the immune system (9). The previous one is done using antibiotics (6). During the course of the infections, the patient needs to be put on a course of antibiotics. Patients with PID may need a longer course that normal patients. If the infection does not respond to the treatment, the patient may need to be hospitalized and put on intravenous antibiotics. After the treatment has been successfully done, some people need to continue take antibiotics as prophylaxis to prevent damage caused by respiratory and ear infections (9).

Boosting the immune system is the actual maintenance therapy (9). Different types of immune therapy are used depending on which compound in the immune system is defective. If the body is having problem producing antibodies, then immunoglobulin therapy is used. The treatment consists of administration of immunoglobulins intravenously or subcutaneously (9).

The goal of treatment with immunoglobulins is for the individual to be free of infection. Today, patients can self-administer immunoglobulins at home, both subcutaneously and intramuscularly. The treatment is also available as intravenous injections. Every other year, the patients go on follow-ups to check immunoglobulin levels and to adjust doses of the treatment depending on the conditions (6).

National Quality Registry for Primary Immunodeficiency

The National Quality Registry for primary immunodeficiency, also called PIDcare, is a registry but also a platform that is used as a decision support tool for healthcare in treatment of individuals with PID. The platform is used by both patients and healthcare (10). The patients can access the platform and fill in daily updates on their current conditions of PID, including symptoms diaries, usage of antibiotics and how they are feeling generally. This information can then be accessed by healthcare as part of the patient’s follow-up (11). The platform is also used by healthcare where medical

examinations and treatments are logged in and can be seen over time, see figure 1. This is a help for healthcare to have to have an overview of how the treatment for this patient has progressed (12).

(8)

Figure 1: an illustration how patent’s information looks like for healthcare personnel. Information on IgG values, antibiotics and bacterial cultures are all included. Other information such as symptoms, infection diagnosis and vaccinations can also be included.

Challenges in modern treatment of immunodeficiency

With modern technique, treatment of PID with immunoglobulins is now done at home instead on the patient having to the hospital to get each dose. In the past, patient

received their dose of immunoglobulins by healthcare. With the patient coming in every other week, healthcare could easily monitor the levels of immunoglobulins and

adjustments to the dose were made accordingly. Also, nurses could talk to the patient about any recent symptoms that might be related to the disease. Since the modern treatment has taken place, the patients have been able to take care of their treatment by themselves in their own homes. However, healthcare does not get to meet the patient as often as before and the time between the follow-ups has become much longer. As a result, the levels of immunoglobulins and infection symptoms cannot be monitored in the same way as before.

As the patient is getting the immunoglobulin replacement therapy at home, it is important for healthcare to know how the treatment is going. One way of doing that is for the patient to fill in a symptom diary, preferably an online one that healthcare can get access to at any time. This way, healthcare gets information about whether the treatment is going well or if the patient is getting infections too often. The information

(9)

from these diaries are then put into PIDcare. One of the most common problems in this case is that the patients do not fill in the diary as they are supposed to. This could affect the design of a study because when the information is being gathered from the registry only, one is relying on the fact that the patient has filled in all the information that is required of them. This is not always achievable since the patients may not always find the time and energy to do write their current conditions down in detail.

Another problem is that the patients’ medical records data bases between different hospitals are not connected. This causes a communication issue between primary care and healthcare. An example of this is when a patient with diagnosed PID gets an infection. According to the Swedish healthcare system, the patient must go to a doctor at the primary care who acts as a gatekeeper versus hospital care. This doctor will then prescribe an antibiotic. The problem here is that this primary care doctor might not see in the patient’s health record that this individual has PID or that the doctor does not have detailed knowledge about the disease. This may result in the patient getting frequent infections without the doctors at the primary care reacting to it. This communication issue could be dangerous for the patient because in these cases, the problem might lay in the immunoglobulin treatment being less efficient. But if the patient only sees primary care, the levels of immunoglobulins might not be checked as they should have been.

One theory to overcome this patient-healthcare and primary care-healthcare miscommunication is to build an alerting system. The idea of this alert system is to launch a signal at healthcare if a patient is being prescribed antibiotics too often. The idea is to classify antibiotics and build an index where every group of antibiotics is given a specific value based on the efficacy and the broadness of its spectrum. This way, even if the patient does not fill in the diary as they are supposed to, the system sends an alerting signal to healthcare in case it suspects that the treatment becoming less efficient.

National Prescription Database

National Prescription Database (NPD) provides the basis for the official statistics on medicines in Sweden and contains information on all the collected prescriptions for every citizen in the country. It also include information on collected eligible

consumable items, such as ostomy products and foods for special nutrition for children under 16 years (13).

Since the registry contains all the collected prescriptions, it contains also all the information that is included in each prescription (13), see table 1.

Table 1: information that is included in a prescription (13).

Information on the patient Name, social security number

Information on the prescribed drug Drug name, concentration or amount of substance in formulation, administration form,

(10)

Box for free text Dosage, usage, purpose of the treatment, duration of the treatment

Information on high-cost protection Amount the patient has spent om medicine, amount left to free-card

Interval Start and end of interval (usually used

only for narcotics)

It should be noted that the dosage described in table 1 is usually given in the box for free text. This may cause an issue because there is no specific box with structured information on the dosage and therefore the correct dosages are not easily calculated. Specific algorithms need to be applied for any system to be able to use the collected data on the dosage from the free-text box in any qualitative way.

The future of healthcare technology

Healthcare technology is always developing and the use of algorithms for important tasks is increasing in the healthcare sector. Algorithms are computerized systems that serves as a solver to one or more problems (14). In healthcare sector, algorithms can be designed to analyze, but also to identify specific individual such as tracking patients with PID (15). This algorithm might be of big help, since many new cases of PID, or worsening of a known PID, can go undiagnosed because of the unspecific symptoms (16).

Another step towards advanced healthcare technology, is the development of artificial intelligence (AI). AI is a branch of computer science focusing on creating smart machines or codes that are capable of preforming tasks that normally require human intelligence (17). Machine learning is another technique where the machine is

continuously using new data and statistics to progressively get better at the given task (17). An example of an approach is using machine learning techniques to identify important genes for disease classification (18). In a study by Keerthikumar and colleagues, candidate PID genes were identified using features of PID genes and non-PID genes. This could be helpful in testing patient where none of the known non-PID genes are involved in their etiology (19).

Ethical aspects of the study

Data in the NPD and PIDcare are covered with absolute confidentiality since they contain personal information on the patients and their treatments. It is possible to extract this kind of data to use them for research purposes (20). In order to get access to such data, an application must be sent to Swedish Ethical Review Authority (SERA) first (21). When the approval has been obtained, an application can be sent to the holder of the desired registry (20).

However, there is an exception that apply to data used in statistical studies. Getting access to these data does not require an approval from SERA. These data are

(11)

individual (22). Since we did not have enough time to obtain an approval from SERA, we framed our study and classified it as a statistical one, so the data could be disclosed without an approval.

The aim of the study

The aim of the study was to develop a predictive Antibiotic Burden Index based on two models; one on number of antibiotic prescriptions fulfilled and the second based on ranks of antibiotics plus number of antibiotic prescriptions fulfilled, to explore whether any of these models can describe how patients’ antibiotic use is related to levels of IgG. The hypothesis was that the antibiotic use will be higher during the six months period prior to the patients lowest recorded IgG-level since the disease condition is then at its worst. In the study this was compared to the antibiotic use six months prior till the highest recorded IgG where the disease condition should have improved. This would tell if the antibiotic use is correlated to the IgG-levels and if ABI then could be used to warn healthcare of a worsening disease.

Methods

Outcome measures

In association with department of infections and immunodeficiency at Huddinge Hospital, outcome measures that could be used in this study was discussed. As mentioned earlier, the goal of the treatment is to lower the rate of infections, and thereby decreasing usage of antibiotics. The outcome measures that should be used are the ones that apply to both types of immunodeficiency that are going to be studies, XLA and CVID. Possible measures could be the lung function described as FEV1, plasma concentration of IgG, aggravation of symptoms and number of bacterial cultures. The FEV1-value is used as a measure of the lung function, which gives a reflection of how much historic infections have affected the lungs. If the infections have been frequently recurrent, the FEV1-value is lower. This outcome measure is useful because it gives a clear picture of how often the lung is infected. On the other hand, patients with immunodeficiency are required to do FEV1-test every 3-5 years, which is too rarely to be used as an outcome measure.

Aggravation of symptoms can be used if the patients are good at documenting their symptoms in their symptom diary. Unfortunately, these symptom diaries are not always being filled as they should, hence it is not well documented in the registry. The same applies to bacterial cultures. The transfer of information from the bacterial culture to the registry is not up to date. Therefore, it was decided that none of these outcome measures could not be used.

Since both XLA and CVID are characterized by deficiency of IgG, this outcome measure could be used. IgG levels are usually measured on the follow-ups of patients with PID and the values are automatically imported into PIDcare. Therefore, this outcome measure was the most suitable one to use.

(12)

The process of obtaining the data

To obtain information on the antibiotics that were used by the studied patient group, an application was sent to the National Board of Health and Welfare (NBHW)

(Socialstyrelsen) which is the holder of NPD in Sweden. First, data on IgG-levels and the social security numbers of each patient in the studied group were extracted from PIDcare. This was done by my supervisor who had the approvals needed to manage this kind of sensitive information. Two dates were also extracted from PIDcare, the date where the patient had his or her lowest IgG (L-IgG) and the date of the highest recorded IgG (H-IgG) and the actual level of IgG from each. Then, these were taken the NBHW on an encrypted USB memory stick, by my supervisor, since the data contains personal information on the patients. The security of these data is very important. From the NPD, information on patients’ antibiotic collection from pharmacies were provided including which class of antibiotic the patient has been prescribed and the date of the collection from the pharmacy. The personal information about the patients were also deidentified by the NBHW and each patient was given a serial number instead. This made the patients untraceable. Unfortunately, information on the dosage and the number of treatment days could not be provided since there is no specific box for them in the electronic prescription, see section National Prescription Database.

Developing the Antibiotic Burden Index

To give a value to the patient’s total use of antibiotic over a specific period of time, Antibiotic Burden Index (ABI) was developed. In the first model, ABI was a sum of values where every antibiotic collected from the pharmacy counts as 1 in the calculation of the index. In the second model each antibiotic class was given a rank, this rank was then used as a part of the ABI calculations. The ranks were applied on each antibiotic according to the efficacy, broadness of the spectrum and the clinical use of them. For instance, doxycycline has a higher rank than penicillin since they are more efficient in many ways and therefore used to treat more serious disease. The valuation of the antibiotic groups was made in collaboration with Dr. Peter Bergman and nurse Susanne Hansen, PID experts at n76 Division of Infectious Diseases and Dermatology, Huddinge Hospital. With their help, we could also get a clinical aspect on the use of these

antibiotics when valuating.

To calculate ABI the patient’s L-IgG and H-IgG was needed. ABI was calculated using data on antibiotics that were collected during the six months period prior to L-IgG and H-IgG respectively, see figure 2. The reason for choosing a period of six months is so that there is enough data to be analyzed while remaining in the timeframe for the project. No consideration on number of treatment days was done since we have no information on exact amount, see the discussion on dosage in the section National

Prescription Database. But according to Dr. Bergman, the days of treatment for the different antibiotic groups doesn’t differ significantly.

(13)

Figure 2: an illustration of the study model. The antibiotic collected from the pharmacy during the six months period up to the L-IgG and H-IgG respectively for each patient, were provided by NBHW. ABI was then calculated according to the two models.

When determining IgG-levels used in this study, Dr. Bergman put forward three potential limits, < 300 mg/dL, < 700 mg/dL and >699 mg/dL. It was decided that two groups should be looked into, the group that had a significantly lower IgG levels, L-IgG < 300 mg/dL, and a larger group containing all the patients that had L-IgG < 700 mg/dL. Furthermore, for the patient to be included in this study, their H-IgG should be > 699 mg/dL, meaning that their condition has significantly improved. Table 2 shows similar limits that has been used in other studies.

Table 2: Approximate limit for the levels of IgG suggested in some studies including a review by the American Primary Immunodeficiency Committee (23–25).

Level of IgG Severity

< 100 mg/dL Profoundly reduced

100 – 300 mg/dL Significant

400-600 mg/dL Adequate

> 500 mg/dL Reduced infections

(14)

Interpretation of data

Since this study was done by collaboration with Huddinge Hospital, the clinic at the hospital was interested in the results concerning Stockholm County. Moreover, results from the rest of the country (non-Stockholm regions) was also interesting to investigate and to compare to local results Finally, a total value of the whole country is also presented. For this reason, there are a total of six groups included in this study; two groups in each regional selection, the ones with L-IgG < 300 mg/dL and a larger group with L-IgG < 700 mg/dL. It should be noted that the data on the patients’ antibiotic use in Stockholm County only include those with CVID and XLA, while the data on patients from the non-Stockholm regions include all PID categories.

For each of these groups, two ABI values were calculated. The first is based on number of collected prescriptions only, i.e. for each time the patient collects an antibiotic from the pharmacy, the score increases with one point. In this case, the type of the antibiotic is not being considered. In the second case, ABI calculations are done based on the ranks of the antibiotics as well as numbers of times it was collected. A valuation of the antibiotic groups was suggested by Dr. Bergman and colleagues where every antibiotic class equals a certain number of points, see table 9. In this case, the rank of the

antibiotic is multiplied by the number of times that specific antibiotic was collected from the pharmacy, during the six months prior to L-IgG and H-IgG respectively. These two models generate two different ABI values for the same patient.

The next step includes calculations of the difference between ABI during the six moths prior to L-IgG compared to ABI during six months prior to H-IgG. To obtain the percentage difference, a total ABI score in each group is divided by number of patients, hence getting the score per capita. Then the score per capita for H-IgG is divided with score per capita for L-IgG. This results in a parentage difference that can be used to draw a conclusion.

Results

ABI based on number of collected antibiotics – model 1

All the ABIs in table 3-8 are calculated for six months prior to L-IgG and H-IgG

respectively. The score per capita is calculated using the number of patients. In addition, a percentage difference between the scores per capita is given

Table 3: ABI data on total patients in Sweden based on number of prescriptions. Patients included in this table had the lowest measured IgG < 300 mg/dL and the highest measured IgG > 699 mg/dL.

ABI score Number of patients Score per capita

ABI prior to L-IgG 304 236 1.288

ABI prior to H-IgG 1544 1490 1.036

(15)

Percentage difference 22.2 %

Table 3 and 4 contains the summed data from the whole country, including Stockholm’s region. As mentioned earlier data from the rest of the country does not only include CVID and XLA, but also other categories within PID.

Table 5: ABI data on patients in non-Stockholm regions, based on number of prescriptions. Patients included in this table had the lowest measured IgG < 300 mg/dL and the highest measured IgG > 699 mg/dL.

ABI

score Number of patients Score per capita

ABI prior to

H-IgG 1375 1393 0.987

Table 6: ABI data on patients in non-Stockholm regions, based on number of prescriptions. Patients included in this table had the lowest measured IgG < 700 mg/dL and the highest measured IgG > 699 mg/dL.

ABI

Table 5 and 6 shows that, unlike Stockholm County, the non-Stockholm regions got a higher ABI six months prior to L-IgG than H-IgG. A mentioned earlier, data from the non-Stockholm regions does not only include CVID and XLA, but also other categories within PID.

(16)

Table 7: ABI data on patients in Stockholm County based on number of prescriptions. Patients included in this table had the lowest measured IgG < 300 mg/dL and the highest measured IgG > 699 mg/dL. The minus in the percentage difference indicated that the result is contrariwise to what was expected.

ABI

Percentage difference -124.0 %

Table 8: ABI data on patients in Stockholm County based on number of prescriptions. Patients included in this table had the lowest measured IgG < 700 mg/dL and the highest measured IgG > 699 mg/dL. The minus in the percentage difference indicated that the result is contrariwise to what was expected.

ABI

Table 7 and 8 contain data on XLA and CVID patients in Stockholm County. The results show that the ABI six months prior to the H-IgG got a higher score than six months prior to L-IgG. It should be noted that the average difference in number of days between L-IgG and H-IgG in Stockholm is 3256 days (approx. 9 years) compared to non-Stockholm regions which is 1298 days (approx. 3,5 years).

(17)

Figure 3: a summarized illustation of the presentage difference between ABI six months prior to L-IgG and six months prior to H-IgG, in model 1. The blue color repersent the groups that had L-IgG < 300 mg/dL and the orange ones the groups that had L-IgG < 700 mg/dL. The size of the circle indicated the relative number of patients.

Looing at the summerized results in figure 3, the results for both studied levels of IgG only show a small percentage difference between them. Sweden in total and

non-Stockholm regions both showed a positive change between ABI prior to L-IgG and ABI prior to H-IgG. Sockholm on the other hand, the results seems to be the opposite. It should be noted however, that the number of patients from Stockholm’s regions are only 97 while non-Stockholm regions are 1393.

ABI based on ranks of antibiotics – model 2

Table 9: The valuation of the antibiotic groups used in PID. The table is provided by Dr. Bergman and colleague at Huddinge Hospital.

Antibiotic class

ATC-number

Example of substance Ranking β-lactamase-sensitive antibiotics J01CE benzylpenicillin 1 β-lactamase-resistant antibiotics J01CF flucloxacillin 1 Penicillin with extended spectrum J01CA amoxicillin 2

Tetracyclines J01AA doxycycline 2

Inhibitor-penicillin combinations J01CR amoxicillin & clavulanic acid 2

Cephalosporins J01DB cefadroxil 3

Fluoroquinolones J01MA ciprofloxacin 3

Sulfonamide and trimethoprim combinations

J01EA sulfamethoxazole, Trimethoprim 3

Macrolides J01FA erythromycin 3

Antibacterial aminoglycosides J01GB tobramycin 4

Carbapenems J01DH meropenem 4

Rifampicin J04AB rifampicin 4

Polymyxins J01XB tadim 4

To be able to calculate ABI for each patient, a ranking of the antibiotic groups is needed. Table 9 shows the valuation that was provided by Dr. Bergman and colleagues based on clinical experience. The lowest rank is given to the penicillins followed by tetracyclines, cephalosporins, antibacterial glycosides and finally high ranked antibiotics that are used in special cases.

In the same way as the results based on number of collected prescriptions, all the ABIs in table 10-15 are calculated for six months prior to L-IgG and H-IgG respectively. The

(18)

score per capita is calculated using the number of patients. In addition, a percentage difference between the scores per capita is given

ABI

Table 10 and 11 contains the summed data from the whole country, including Stockholm’s region. As mentioned earlier data from the rest of the country does not only include CVID and XLA, but also other categories within PID.

Table 12: ABI data on patients in non-Stockholm regions, based on ranks of antibiotics. Patients included in this table had the lowest measured IgG < 300 mg/dL and the highest measured IgG > 699 mg/dL.

ABI

Table 13: ABI data on patients in non-Stockholm regions, based on ranks of antibiotics. Patients included in this table had the lowest measured IgG < 700 mg/dL and the highest measured IgG > 699 mg/dL.

(19)

Table 12 and 13 shows that, unlike Stockholm County the non-Stockholm regions got a higher ABI six months prior to L-IgG than H-IgG. As noted earlier, data from the non-Stockholm regions does not only include CVID and XLA, but also other categories within PID.

Table 14: ABI data on patients in Stockholm County based on ranks of antibiotics. Patients included in this table had the lowest measured IgG < 300 mg/dL and the highest measured IgG > 699 mg/dL. The minus in the percentage difference indicated that the result is contrariwise to what was expected.

Table 15: ABI data on patients in Stockholm County based on ranks of antibiotics. Patients included in this table had the lowest measured IgG < 700 mg/dL and the highest measured IgG > 699 mg/dL. The minus in the percentage difference indicated that the result is contrariwise to what was expected.

Table 14 and 15 contain data on XLA and CVID patients in Stockholm County. The results show that the ABI six months prior to the H-IgG got a higher score than six months prior to L-IgG. As mentioned earlier, the average difference in number of days between L-IgG and H-IgG in Stockholm is approx. 9 years compared to non-Stockholm regions which is approx. 3,5 years.

(20)

Figure 4: a summarized illustation of the presentage difference between ABI six months prior to L-IgG and six months prior to H-IgG, in model 2. The blue color repersent the groups that had L-IgG < 300 mg/dL and the orange ones the groups that had L-IgG < 700 mg/dL. The size of the circle indicated the relative number of patients.

Looing at the summerized results in figure 4, the results for both studied levels of IgG only show a small percentage difference between them. Sweden in total and

non-Stockholm regions both showed a positive change between ABI prior to L-IgG and ABI prior to H-IgG. Sockholm on the other hand, the results seems to be the opposite. It should be noted however, that the number of patients from Stockholm’s regions are only 97 while non-Stockholm regions are 1393.

Discussion

In general, no difference in pattern was discovered between using model 1 and model 2. As described in figure 2, ABI is calculated from data six months up to L-IgG and H-IgG respectively. The percentage numbers may differ between figure 3 and 4 but the pattern is the same. The advantage of using the antibiotic ranking system is that the efficacy of the antibiotic is taken into account. This may give a more correct picture of how the patient is doing and the status of the disease. Other variables such as dose of the drug and treatment length is also valuable information. The problem here is that the Swedish prescription system doesn’t include structured data on dosage and treatment length, which makes it difficult to extract this kind of information (13), see National

Prescription Database. The dose of the drug and treatment length is given as free text,

thus the need of complicated algorithms to interpret and extract this kind of data (15). Since this is an explorative study, only simple variables were taken into consideration, but if this project is further developed into a proper functioning warning system, the other variables may need to be included.

ABI based on number of prescriptions was calculated for Stockholm’s region, non-Stockholm regions and Sweden in total. When looking at the summarized results in figure 3, the non-Stockholm regions seem to confirm the hypothesis. The percentage rate shows that there is a difference within each group. In non-Stockholm regions, a

(21)

difference of 34,7 % was presented in the IgG < 300 mg/dL group and a 29,0 % in L-IgG < 700 mg/dL group. The improvement implies that the ABI prior to H-L-IgG is lower, meaning that the patients are using less antibiotics. Meanwhile, in Stockholm’s region the results are contrariwise of what was predicted in the hypothesis. This is strange because this means that the patients are using more antibiotics prior to their H-IgG than L-IgG. The average difference in number of days between L-IgG and H-IgG in

Stockholm equals approx. 9 years compared to non-Stockholm regions which equals approx. 3,5 years. One reason for the deviation of Stockholm’s region’s results could be because of changes in the treatment recommendations in the area during the years that the study material includes. Another reason could be fact that the data from

Stockholm’s region only include XLA and CVID, while non-Stockholm regions include all PID classes. There could be some kind of physiological factor causing this

difference. And since we don’t have an ethics committee approval to obtain exact dates for L-IgG and H-IgG, we were not able to do calculations on date and time intervals to investigate if this could be the reason why Stockholm’s region is deviating from rest of the country. Looking at Sweden in total, the pattern is the same as the non-Stockholm regions. Obviously, it is the deviating values from Stockholm’s region that is causing the Sweden in total percentage to be lower.

In the same manner, when looking at ABI based on ranks of antibiotics, non-Stockholm regions matched the hypothesis while Stockholm’s region is deviates, see figure 3. Since the only difference between the two models is the ranking of the antibiotics, the reason for the deviation should be the same in both cases.

Other than the deviating values from Stockholm’s region, the study seems to have given promising results. Since no statistical analysis was done on the data, the precise

significant difference cannot be presented. However, the percentage values in tables 5-6 and 12-13 indicate that, in non-Stockholm regions, there are differences in ABI during the six months up to L-IgG and H-IgG respectively. This means that with further improvements and inclusion of other crucial variables, a warning system could be feasible. This warning system will facilitate for healthcare to keep an eye on PID patients and will fire a signal if the ABI becomes too high, that would then be an indication of immunoglobulin treatment not being as efficient as it should. In the same manner as the study described earlier by Keerthikumar and colleagues where AI and machine learning were used to detect PID genes, this kind of technologies could also be incorporated to build a sustainable and efficient system for detecting peeks in ABI (19). In such system, a new ABI for the patient will be constantly calculated over a specific period of time backwards for every antibiotic prescribed. In the platform, the data will ideally be in real-time and directly connected to NPD. This way, there will be no need for the healthcare staff to do any transmission of data between the systems. Another benefit of this warning system could be the reduction of the patients’ need for antibiotics hence reducing the risk of antibiotic resistance.

A limitation of this study is that we could not specify exact H-IgG-levels and L-IgG-levels for these patients since we did not have the time to obtain the ethics committee approval to access exact patient data (20). Instead we were forced to use categorize the patients into two groups IgG < 300 mg/dL and IgG < 700 mg/dL. This makes it difficult to specify the exact correlation between IgG-levels and antibiotic use.

(22)

The next step after this study may be a more profound analysis of the model. For instance, taking 25 % of the patients and optimizing the antibiotic ranking system according to those patients to get the highest percentual value possible, describing the improvement. Afterwards, the new developed ranking system could be applied to the rest of the 75 % of the patients and check if the model works on them too. This could be a good start to establish a basis that can be developed into the warning system

mentioned earlier.

Conclusion

Although the results from Stockholm’s region are deviating, both models are showing promising results that can be used to develop a functioned warning system. There seems to be a correlation between the use of antibiotics and IgG-levels. The hypothesis was accurate.

Acknowledgements

I would like to thank my supervisor Joakim Söderberg at Health Solution AB for making this project possible and for all the support and encouragement during the thesis process.

I would also like to thank our collaborators, Dr. Bergman and Susanne Hansen at Huddinge Hospital, for providing the clinical aspects and expertise on PID, needed for this project.

(23)

References

1. Geha RS, Notarangelo LD, Casanova J-L, Chapel H, Conley ME, Fischer A, et al. Primary immunodeficiency diseases: an update from the International Union of Immunological Societies Primary Immunodeficiency Diseases Classification Committee. J Allergy Clin Immunol. 2007 Oct;120(4):776–94.

2. Boyle JM, Buckley RH. Population prevalence of diagnosed primary immunodeficiency diseases in the United States. J Clin Immunol. 2007 Sep;27(5):497–502.

3. Bonilla FA, Bernstein IL, Khan DA, Ballas ZK, Chinen J, Frank MM, et al. Practice parameter for the diagnosis and management of primary

immunodeficiency. Ann Allergy Asthma Immunol. 2005 May;94(5 Suppl 1):S1-63.

4. Expertteam för Primär immunbrist [Internet]. [cited 2019 Oct 25]. Available from: http://www.karolinska.se/for-vardgivare/kliniker-och-enheter-a-o/kliniker-och- enheter-a-o/funktion-karolinska-universitetslaboratoriet/centrum-for-sallsynta-diagnoser/regionala-expertteam/primar-immunbrist/

5. Kiaee F, Azizi G, Rafiemanesh H, Zainaldain H, Sadaat Rizvi F, Alizadeh M, et al. Malignancy in common variable immunodeficiency: a systematic review and meta-analysis. Expert Rev Clin Immunol. 2019 Oct;15(10):1105–13.

6. Variabel immunbrist [Internet]. SITENAME. [cited 2019 Sep 8]. Available from:

https://www.socialstyrelsen.se/stod-i-arbetet/ovanliga-diagnoser/variabel-immunbrist/

7. Reference GH. X-linked agammaglobulinemia [Internet]. Genetics Home Reference. [cited 2019 Oct 25]. Available from:

https://ghr.nlm.nih.gov/condition/x-linked-agammaglobulinemia

8. Smith CE, Berglöf A. X-Linked Agammaglobulinemia. In: Adam MP, Ardinger HH, Pagon RA, Wallace SE, Bean LJ, Stephens K, et al., editors. GeneReviews® [Internet]. Seattle (WA): University of Washington, Seattle; 1993 [cited 2019 Oct 25]. Available from: http://www.ncbi.nlm.nih.gov/books/NBK1453/

9. Primary immunodeficiency - Diagnosis and treatment - Mayo Clinic [Internet]. [cited 2019 Oct 7]. Available from: https://www.mayoclinic.org/diseases-conditions/primary-immunodeficiency/diagnosis-treatment/drc-20376910 10. Om PIDcare – PIDcare [Internet]. [cited 2019 Oct 25]. Available from:

http://pidcare.se/om-pidcare/

11. För patienter och vårdnadshavare – PIDcare [Internet]. [cited 2019 Oct 25]. Available from: http://pidcare.se/for-patienter/

(24)

12. Nyttan med PIDcare och Hälsodagboken – PIDcare [Internet]. [cited 2019 Oct 25]. Available from: http://pidcare.se/nyttan-med-pidcare/

13. Läkemedelsregistret [Internet]. SITENAME. [cited 2019 Oct 25]. Available from:

https://www.socialstyrelsen.se/statistik-och-data/register/alla-register/lakemedelsregistret/

14. O’Mara-Eves A, Thomas J, McNaught J, Miwa M, Ananiadou S. Using text mining for study identification in systematic reviews: a systematic review of current approaches. Systematic Reviews. 2015 Jan 14;4(1):5.

15. Modell V, Gee B, Lewis DB, Orange JS, Roifman CM, Routes JM, et al. Global study of primary immunodeficiency diseases (PI)--diagnosis, treatment, and economic impact: an updated report from the Jeffrey Modell Foundation. Immunol Res. 2011 Oct;51(1):61–70.

16. Chapel HM. Consensus on diagnosis and management of primary antibody deficiencies. Consensus Panel for the Diagnosis and Management of Primary Antibody Deficiencies. BMJ. 1994 Feb 26;308(6928):581–5.

17. What is Artificial Intelligence? How Does AI Work? | Built In [Internet]. [cited 2019 Nov 30]. Available from: https://builtin.com/artificial-intelligence

18. Zhang HH, Ahn J, Lin X, Park C. Gene selection using support vector machines with non-convex penalty. Bioinformatics. 2006 Jan 1;22(1):88–95.

19. Keerthikumar S, Bhadra S, Kandasamy K, Raju R, Ramachandra YL, Bhattacharyya C, et al. Prediction of Candidate Primary Immunodeficiency Disease Genes Using a Support Vector Machine Learning Approach. DNA Res. 2009 Dec;16(6):345–51.

20. Om sekretessen kring register [Internet]. SITENAME. [cited 2020 Feb 23]. Available from: https://www.socialstyrelsen.se/statistik-och-data/register/om-sekretessen/

21. Vanliga frågor [Internet]. Etikprövningsmyndigheten. [cited 2020 Feb 23]. Available from: https://etikprovningsmyndigheten.se/vanliga-fragor/ 22. Register [Internet]. SITENAME. [cited 2020 Feb 23]. Available from:

https://www.socialstyrelsen.se/statistik-och-data/register/

23. Sorensen RU, Moore C. Antibody deficiency syndromes. Pediatr Clin North Am. 2000 Dec;47(6):1225–52.

24. Bonagura VR, Marchlewski R, Cox A, Rosenthal DW. Biologic IgG level in primary immunodeficiency disease: The IgG level that protects against recurrent infection. Journal of Allergy and Clinical Immunology. 2008 Jul 1;122(1):210–2.

(25)

25. Orange JS, Hossny EM, Weiler CR, Ballow M, Berger M, Bonilla FA, et al. Use of intravenous immunoglobulin in human disease: a review of evidence by

members of the Primary Immunodeficiency Committee of the American Academy of Allergy, Asthma and Immunology. J Allergy Clin Immunol. 2006 Apr;117(4 Suppl):S525-553.

Appendix

Appendix A

Figure A-1: an illustration of how patient’s information looks like for healthcare personnel. Infromation on IgG values, antibiotics and bacterial cultures are all included. The red circle illustrated the