Drug Safety Issues Focusing on Liver and Pancreas Toxicity

Background: Medications benefit people with diseases and medical conditions, improving many patients' lives and in some cases significantly increasing their quality of life. Despite these great benefits, medications can lead to serious adverse effects. Adverse drug reactions (ADRs) are the 4th leading cause of death, greater than major killers including pulmonary diseases, diabetes and automobile death. Not all medication groups have the same incidence in inducing diseases. Anti-infective agents including antibacterial agentsare the most likely to induce diseases followed by cardiovascular agents and antineoplastic agents. Objectives: Drug safety is a very important issue in patient therapy. Since antibacterial agents in general and fluoroquinolones specifically are among the most prescribed medications, it is very important to quantify their risks in causing serious adverse reactions such as hepatotoxicity and pancreatitis. In terms of drug safety and preventing ADRs, costs are a very important factor because many conditions like hypoglycemia can be prevented, leading to significant decreases in cost and improvement in quality of life. Since there is a dearth of research that has examined these areas, epidemiological studies are needed. The objectives of the three proposed manuscripts of this dissertation were to a) estimate the risk of hepatotoxicity associated with fluoroquinolones use; b) examine antibacterial agents use and the risk ofacute pancreatitis development; c) review the healthcare databases in United States and European countries that are commonly utilized in conducting epidemiological research. Methods: Two matched case control studies were used to examine the risk of hepatotoxicity and acute pancreatitis development with using fluoroquinolones and antibacterial agents, respectively. The Veterans' Affairs (VA) medical database was used to perform the two studies. Odds ratios (OR) and their 95% confidence intervals (CI) were derived from crude and adjusted conditional logistic regression models. In the third paper, a literature search was performed using Publisher Medline (Pubmed), Embase® and the Iowa Drug Information Services (IDIS) to perform a review of four medical databases including the Nationwide Inpatient Sample database, the Veterans' Affairs medical database, the Health Improvement Network database and the Norwegian medical database. Results: In the first study, fluoroquinolones use was associated with increased risk of hepatotoxicity. However, only ciprofloxacin was statistically significantly associated with development of hepatotoxicity while levofloxacin and moxifloxacin were not associated with hepatotoxicity development. In the second study, use of antibacterial agents was associated with increased risk of acute pancreatitis. Conclusion: The findings fro in both studies show that antibacterial agents are associated with increased risk of hepatotoxicity and acute pancreatitis. Knowing the safety of medication is very important for the clinical practice especially for antibacterial agents since they are commonly used and clinicians are considering these agents very safe. However, further research is needed to confirm these findings and to understand the biological mechanism behind their toxicity.

After I arrived here in the United States, I never felt that I was away from my family because I felt as though I was part of another great family, which includes my professors and fellow graduate students.
First, I would like to thank Prof Paul Larrat for his encouragement and help during my studying at URI. Thanks to him for believing in me and taking the time to guide and keep me on the track. He has been very supportive and always made it easy for me with his kind words of encouragement and the humble heart that he has.
Thanks a lot for his advice, efforts, time and everything that he has done for me. He truly has had a great influence on my personality and my success.
iv I would like to thank Dr.Kerry LaPlante for helping me to master a great database. I remembered first time I met her, she was so kind and very motivated to help me. Although she was so busy, she gave me her attention and advice every time I contacted her. She was very supportive and always encouraged me to do more because she believes in me. I was so lucky to work on the VA database and without her help this dissertation would not have been possible. He was so motivated to help me as one of his student in his class to understand the statistical method that is related to my projects.
I would like to thank Dr.Stephen Kogut. I can't express my feelings towards him. Dr .Ko gut dealt with me as his brother. I had the honor of meeting his great family and I was so happy to have this chance. As a professor, Dr.Kogut supported me during the first semester here at URI. His office was always opened for me. He gave me from his time no matter how busy he was. I was lucky to take many classes with v List of Tables Chapter I   Table 1

Background
Hepatotoxicity, defined as an injury to the liver cells is commonly associated with a decrease in liver function.
1 The liver's metabolism and connection with the gastro-intestinal tract increases susceptibility to injury from drugs and other ingested substances. Hepatotoxicity can range from a mild reaction, which can be resolved by discontinuation of the causative agents to a serious fatal reaction. 2 Medications, herbs, chemicals and infectious pathogens are the most common causes of hepatotoxicity. 3 -7 Although the incidence of drug induced-hepatotoxicity is considered rare -1 in 10,000 to 1 in 100,000-it is usually clinically significant. 8 Fifty percent of acute liver failure cases are caused by medications and 5% of all hospital admissions are caused by drug-related hepatotoxicity. 9 Currently, hepatotoxicity is the leading cause of acute liver failure in the United States (US).
Acute Liver Failure (ALF) due to medications is usually a fatal reaction and only 20% of patients will respond to therapy and survive.2 Liver damage due to drugs has different features: hepatocellular, cholestatic or a.mixture ofboth. 10 - 12 There is no specific treatment for drug-induced hepatotoxicity in most cases; supportive care and withdrawal of causative agent are the only treatment options available. More than 900 drugs can cause hepatotoxicity. 13 As a cost ofliver toxicity, the average cost of treating a patient with acute liver failure using 2001 and 2006 data is estimated at is $146,972 to $252, 113. 14 , 15 Fluoroquinolones are antimicrobial agents that possess a broad spectrum of activity against gram-negative and gram-positive organisms. These drugs are widely "b d d" . 17 one of the most prescn e me 1cat1ons.
Antimicrobial agents, including fluoroquinolones are the most common cause of non-acetaminophen induced hepatotoxicity. 18 Antimicrobial induced hepatotoxicity usually is an idiosyncratic reaction, dose independent and unrelated to intended pharmacological effect. 18 ' 19 In the past decade, liver toxicity has been one of the most common reasons for drug removal from the U.S. market by the FDA such as troglitazone (Rezulin®). 18 In the United Kingdom (UK), the Medicines and Healthcare products Regulatory Agency (MHRA) restricted the indication of moxifloxacin for use only when other antibiotics failed to treat acute bacterial sinusitis, acute exacerbation of chronic bronchitis or community acquired pneumonia due to the risk of hepatotoxicity associated with moxifloxacin. 20 Increases in the reports ofhepatotoxicity associated with the usage of fluoroquino lones are a crucial issue in clinical practice, since fluoroquino lones are commonly prescribed because of their broad spectrum of coverage. At the same time liver toxicity is a serious issue and 50% of cases are related to medication and not related to any underlying diseases. However, in the case reports, the association between fluoroquinolones and hepatotoxicity is unclear. 21 -24 A well designed epidemiological study is necessary to assess the possible increased risk of hepatotoxicity associated with fluoroquinolones usage. Therefore, the purpose of this 5 study was to conduct a real-world safety study to assess the association between fluoroquinolones use and risk ofhepatotoxicity. 6

Data source
In United States, the Veterans Affairs (VA) medical database is considered from the largest medical database with information for more than 7.9 million veterans.
2s- 2 7 These data include detailed information on all medical and pharmacy information for each patient. All patient data are stored in electronic form. code system is used to code some procedures in the medical SAS outpatient datasets. 31 The Diseases coding in the VA database h~s been validated and determined to be of high quality. 30 , 32 Study design A matched case-control study was utilized using a national cohort of patients in the VA databases (hospital and pharmacy data) which contain preexisting data to examine the association between fluoroquinolone use and development of 7 hepatotoxicity. Scrambled Social Security numbers were used to guarantee the privacy of the patients. The study was approved by the Institutional Review Board (IRB) of the y A research committee and the research office at the University of Rhode Island.

Cases and controls
Our patient cohorts were taken from the national VA database in Austin

Discussion
To our knowledge, this is the first study to examine the association between fluoroquinolone antibiotic use and the risk of developing hepatotoxicity. This study utilized an extremely large database. 11 Our population was mainly white and male as expected since most of the patients were veterans. Patients with hepatotoxicity in our study were younger than the controls. This is consistent with many case reports and studies that assess the association between hepatotoxicity and medications.  43 44 h h reported elevated enzymes wit uoroqumo ones use. ' In umans, t e intrahepatic concentrations of fluoroquinolones have been found to be eight times as their concentration in the serum. 45 Increasing the concentration of fluoroquinolones in the liver might lead to hepatic injury. However, the liver toxicity usually occurred when the concentration is 20 times intrahepatically compared to serum concentration. 45 Therefore, this may be the reason why ciprofloxacin is associated with hepatotoxicity and levofloxacin was not.  48 In addition to the metabolism effect mechanism, some literature suggested that fluoroquinolones hepatotoxicity is a idiosyncratic drug reaction "immune-mediated" manifested by eosinophilia. 49 This mechanism could be the suggested mechanism in the case of medication that is mainly eliminated by renal routes such as levofloxacin.
Although, there is no study to assess the association between fluoroquinolones use and risk of hepatotoxicity, SB Meropol et al, compared the incidence rate of hepatotoxicity with using three different medications which are ciprofloxacin , 14 amoxicillin and doxycycline. 50 They compared the incidence of hepatotoxicity with these medications in three different databases. In our study, we controlled for covariates (social behavior, health conditions and medications) that might influence hepatotoxicity. Further, we controlled for the possible interactions (2-way and 3-way) to minimize the role of effect modifiers.
Observational studies, especially those that assess the effects of medications, should evaluate the potential of confounding by indication, when seriously ill patients being more likely to receive fluoroquinolone medications and more likely to develop the outcome. To assess this, first we excluded any patient who has been admitted in the hospital to avoid including any very ill patient. Additionally, we were able to exclude many major risk factor for hepatotoxicity such as viral hepatitis and liver abscess and finally we controlled for most of relevant comorbidities in our regression model.
The recognized association between fluoroquino lones and hepatotoxicity is not new. Trovafloxacin was the first drug from fluoroquinolone medications that associated with hepatotoxicity. 63 Trovafloxacin is available with special restrictions for use only in patients with life-threatening or limb infections because of serious liver toxicity including hepatitis, liver failure that resulted in liver transplantation and death. 64 , 65 In 1999, trovfloxacin was removed from European market because of liver • • 66 tox1c1ty.
Our study has several strengths; we utilized a large database with sufficient power to assess the association between fluoroquinolones use and risk of developing hepatotoxicity. Moreover, this database allowed us to include patients that may not be in other databases, such as those of Native American race and users of illicit drugs.
We were able to control for many major potential confounders such as liver cancer, liver cirrhosis and other risk factors.
However, our study has several limitations. Since we are using an administrative database, there is potentially a problem with coding and missing data, especially for behavioral variables such as alcohol abuse and tobacco smoking. In the VA database, there are specific variables for alcohol abuse and tobacco use but these variables usually are missing at least in the case of our study. Instead, we used ICD-9-CM codes to detect patients who are smoking tobacco, who have history of smoking and patients who are abuse alcoho I. We believe that there may be a an underestimation of the prevalence of these behaviors. Additionally, illicit drug use is determined by patients (self identified), not by using analytical test to test these substances in the blood. Using this method could underestimate illicit drug use and we speculate that not all patients will identify if they use illicit drugs or not. The validity of the ICD-9-CM code as an indicator for hepatotoxicity is not known. In general, the validity of different ICD-9 codes varies.

'
68 However, it is unlikely that hepatotoxicity is ignored as a diagnosis. Another limitation of using administrative database is that we can't determine whether patients receiving fluoroquinolones actually took their medications and adhered to their medication regimen .. The majority of our population in the study are veterans males, therefore we may not generalize our results to women and non-veteran patients.
Finally, as with other observational studies, residual confounding cannot be completely excluded. 18

Conclusion
In conclusion, it was found in a large national cohort study that fluoroquinolones use is significantly associated with the risk of developing hepatotoxicity compared to nonuse. Furthermore, ciprofloxacin was statistically significant associated with the risk of hepatotoxicity compared to non-users. At present, it is believed that clinicians should be aware of this potential issue when they prescribe fluoroquinolones in patients who are at risk of developing hepatotoxicity. Thus, further research is needed to confirm these findings and to better understand the underlying mechanism.

Background and Clinical Significance
Acute pancreatitis is an inflammatory disease that affects the pancreas. It usually affects the pancreatic tissues and in some cases, the inflammation involves the surrounding tissues. I The amount of inflammation will vary in severity from mild to severe, to life threatening. Most cases are mild, resolve within days and may only require a short hospital stay. Severe cases may require hospitalized for days and sometimes for months due to organ dysfunction. In addition to variety in severity, acute pancreatitis can occur as a single attack or it can also be recurrent. Acute pancreatitis is associated with 15-35% mortality rate and the incidence of acute pancreatitis has increased worldwide. The concern over medication, specifically antibacterial agents, causing acute pancreatitis is important because of the popularity of these agents. Antibacterial agents are used in clinical practice in both community and hospital settings. There is therefore a need to understand and determine these potential medical problems risks of acute pancreatitis associated with antibacterial medications. 23 ' 29 There are currently a series of epidemiological studies that are assessing the risk of acute pancreatitis with different groups of medication (e.g. statins, ACEis, d . is 23 30 diuretics). These stu 1es pre 1ct t ens w1 contmue to mcrease. ' ' There are few studies that assessed the risk of acute pancreatitis, particularly with antibacterial agents. One study assessed diseases and medications as risk factors of acute pancreatitis. 7 another study investigated the risk of acute pancreatitis with use of metronidazole. 31 The aim of the study is to assess and evaluate the risk of acute pancreatitis with use of antibacterial drugs among a national cohort of patients Statistical Analyses. The main analysis of the current study was to investigate the association between the risk of acute pancreatitis and exposure to antibacterial medications. A chi-square test or Fisher exact tests were used to analyze the categorical data while a student's t-test or Mann Whitney test were used to analyze the Continuous variables for both demographic and clinical data. Odds ratio and their 95% confidence interval (CI) were derived from the conditional logistic regression.
Initially, the potential clinical covariates that would be entered the multivariable model were selected based on bivariate analyses. Covariates with differences of 10% between the cases and controls were included in the multivariable model. Then, the final model was built using stepwise forward entry method (non-computer generated).
At each time, a covariate was added, and ifthe estimated coefficient (B) of antimicrobial exposure changed by more than 10%, this covariate was kept in the final model. Furthermore, two-way and three-way interactions were assessed. At each step, an interaction term was added to the model, and if the estimated coefficient (B) of antimicrobial exposure changed by more than 10%, the interaction term was kept in the model. Multicollinearity was assessed using variance inflation factor (VIF). A cutoff of2.5 for VIF was used to indicate the presence of multicollinearity. All statistical tests were conducted with a two-tailed alpha of 0.05. All analyses were conducted using SAS® version 9 .1.3 (SAS® Institute, Inc, Cary, NC).

Results
Between January 1, 2002 and June 1, 2009, a total of23,899 patients were identified with a primary diagnosis of acute pancreatitis. Inclusion and exclusion criterion for the study were applied to these patients. Among these cases, patients with of any type of cancer, including pancreatic cancer, (n=899), chronic pancreatitis (n= 2,861), HIV or AIDS (n=369), bacteremia (n=578), ERCP (n=295), abdominal injury (n=9), alcohol related diseases (n=l,842), .gall bladder or biliary diseases (n=3,528), peptic/hemorrhagic ulcer or gastritis (n=376) and esophagitis (n=2,222) were excluded. In addition, patients who were admitted to the hospital for any medical conditions in the previous 180 days prior to the index date (n=718) were excluded from the study. After application of the exclusion criteria, 10,202 patients were eligible for the analyses.
The Distribution of demographic and clinical characteristics of the cases (patients with acute pancreatitis) and the controls (patients with acute myocardial infarction) can be found in Table 11.2. Patients in the case group were younger (mean age 55.6) than the patients in the control group (mean age 64.4). The majority of our population was male. In general, our sample's racial breakdown was mainly white non-Hispanic (58%), followed by black (18.5%). The control group had a higher percentage of white non-Hispanic (63.9%) compared to the case group (52.3%). The case group had a higher percentage of black patients (24.3%) compared to the control group (12.6%). Also, there were more native Americans in the case group (0.50%) compared to the control group (0.30%). There was no difference among Asian and Hispanic races.

41
The case group differed from the control group in alcohol abuse, hypertriglyceridemia, hypercalcemia, and IBD, while there was no difference in other covariates such as diabetes mellitus and use of comparable prescription medications ( Table II.2).
Fluoroquinolones antibiotics were the most used antibiotics among case group patients and control group patients (2.8% and 2.9 respectively), followed by penicillin (2.7% and 2.3%, respectively). Nitrofurantoin (0.03 and 0.08, respectively) and antituberculoses medication (0.04 and 0.05, respectively) were the lowest utilized antibiotics by the study population. Use of sulfamethoxazole/trimethoprim was equal among case group patients and control group patients (0.9%).
Odds ratios (OR) and the 95% confidence intervals (CI) were computed from the adjusted conditional logistic regression model (Table II.3 & Table II.4). After controlling for all of the potential confounders in the multivariable model, there was a statistically significant association between antibacterial use and developing acute pancreatitis. The adjusted odds ratio (AOR) was 1.29 (95% CI 1.18 -1.41 ), suggesting that patients with acute pancreatitis are 1.29 times more likely to have been exposed to antibacterial agents than control group.

Discussion
The current study utilized a matched case control study design using the national Veterans' Affairs database to investigate the association between the use of antibacterial drugs and acute pancreatitis development. To current knowledge, this is the first epidemiological study to assess this association for all oral antibacterial agents as one group in the United States.
There is a known risk with metronidazole use and the risk of acute pancreatitis.
Previous investigations found an association with metronidazole use within thirty days prior to index date (AOR was 3 .0, 95% CI 1.4-6.6). 31 Furthermore, they found that patients who used metronidazole with proton pump inhibitors to treat peptic ulcers due to helicobacter pylori and/or using amoxicillin, macrolides, or tetracycline within thirty days prior to index date were at risk of developing acute pancreatitis (AOR was 8.3, 95% CI 2.6-26.4). 31 However, it is important to note that the increased risk in patients with peptic ulcers could be due other confounding factors.
Investigators in Swede~ used a population case control study to investigate disease and medication risk factors for acute pancreatitis. The risk of using systemic antibacterial agents was evaluated in addition to several other medication classes.
After adjusting all the potential confounders, the AOR was 1.9, 95% CI 1. Another report described a case of acute pancreatitis that resulted from using tigecycline, a compound related to tetracyclines. 57 The patient was a thirty five year old male, who had undergone surgery to remove tissue that was infected with Enterobacter cloacae. He was initially given tigecycline, imipenem and amikacin.
After thirteen days of treatment, he developed upper abdominal pain, and his lipase level was elevated. He was subsequently diagnosed with acute pancreatitis.
Tigecycline and amikacin were stopped; the patient's lipase level then started to decrease and he was no longer experiencing any abdominal pain. After five months, his lab tests returned to baseline.
The incidence of drug induced acute pancreatitis is between 0.1-2%. 19 Although the exact mechanism of antibacterial induced acute pancreatitis is still not known, several mechanisms were suggested for different antibacterial groups. The mechanisms for tetracyclines to be associated with acute pancreatitis were suggested as: (1) accumulation of toxic metabolite, (2) hypertriglycerdemia induced by tetracyclines, and (3) increased concentration of tetracyclines in the biliary duct. It is theorized that the formation and accumulation of a toxic tetracycline metabolite is responsible for causing acute pancreatitis, although no metabolite has been identified 44 yet. 58 Furthermore, tetracyclines inhibit protein synthesis, which could lead to a build up of defective protein within the hepatocyte and inhibit triglyceride release to the liver. This process leads to increased triglyceride levels causing acute pancreatitis. 59 The other possible mechanism is that the high concentration of tetracyclines in the bile is the major reason for tetracyclines induced acute pancreatitis. 57 Minocycline concentration in bile was found to be ten times the concurrent serum concentration. 59 Also, the concentration of doxycycline and tetracycline in the bile reached up to eight times and 1 O times the serum concentration, respectively.

6°
Further, tigecycline biliary elimination rate is 500 times the plasma rate, which could be a major reason for . . 59 acute pancreatlt1s.
An rational for acute pancreatitis and macrolides use has been postulated. It is suggested that macrolides cause acute pancreatitis through their prokinetic property. Antibacterial agents are considered an integral part in the therapeutic plan for acute pancreatitis to prevent the complications of acute pancreatitis such as bacteremia. 1 However, using antibacterial agents in acute pancreatitis still remains controversial. 69 Using antibacterial agents in patients with acute pancreatitis is based upon results from a randomized clinical trial that found administration of imipenem reduced infectious complications. 70 However, a recent randomized clinical trial that compared ciprofloxacin plus metronidazole to placebo and found there is no difference in outcomes between the two groups. 71 Therefore, routine use of prophylactic antibiotics in acute pancreatitis is not recommended until the patient develops an infection. This study's results also support these recommendations since it was found that antibacterial agents were associated with acute pancreatitis. Thus, unnecessarily giving the patient with acute pancreatitis prophylactic antibacterial agents to prevent infections should be avoided since these medications were associated with acute pancreatitis development.
Several indicators were used to find patients with alcohol use among the cases and controls. These indicators included ICD-9 codes, alcohol dependency treatment indicators, and alcohol-related DRGs. This variable in the model was determined using the previous indicators. The Department of Veterans Affairs (VA) requires an annual alcohol screening for any VA patient. 72 In 2000, the screening rate was 85%, which is considering a high rate. 73 To continue this screening program of VA patients, the Veterans Health Administration (VHA) implemented an evidence-based alcohol screening program. A study done in 2006 found that the screening rate of alcohol misuse was 93%. 74 Additionally, the Computerized Patient Record System (CPRS) at the VA has a reminder for the clinician to perform an AUDIT-consumption (AUDIT-C) questionnaire for alcohol misuse. 74 These programs were established for the purpose of screening as well as providing counseling to patients with alcohol misuse.
Based on these programs, it was determined that virtually all VA patients werescreened for alcohol use.
In any observational study, the potential of confounding by indication should be evaluated. Confounding by indication is when patients with severe diseases and comorbidities were more likely to received antibacterial agents than those with moderate diseases. In this study, patients with major risk factors (e.g. patients with gall bladder diseases and alcohol related disease) and those who had been admitted in the hospital during the last six months prior to the index date were excluded from the study. Additionally, the remaining variables were controlled in the regression model.
This study has several strengths. The use of a very large database provides data for a large sample size. This allowed all exclusion and inclusion criteria that have been identified to be applied. Additionally, many potential confounders such as hypertriglycerdemia, alcohol abuse, and other confounders were controlled.
While this study has several strengths, it also has potential limitations. As with all administrative databases, the pharmacy data describes the dispensed antibacterial agents, but it is not known if the patients were actually administered their medications.
Also, the validity of acute pancreatitis diagnosis in the VA databases was not evaluated and assessed in an epidemiological study. The diagnosis of acute pancreatitis may be underestimated and not classified in mild cases. 75 additionally, incideince of drug induced acute pancreatitis is not accurate because it is often misdiagnosed, undiagnosed and underreporting. 19 Furthermore, as with most of 47 administrative databases, missing data and missing codes are always potential problems. In this study, initial plans included incorporating BMI in the analysis.
Weight and height were used to calculate BMI, but up to 48% of the height and weight values were missing. Therefore, it was decided not to include BMI in this study.
Additionally, the majority of the population in the study are male veterans, therefore the results may not be generalized to women and non-veteran patients. Finally, as with other observational studies, residual confounding cannot be completely excluded.

Conclusion
In conclusion, it was found that antibacterial medications were significantly associated with increased risk of acute pancreatitis among a large national sample of veterans. In addition, further studies are required to confirm this conclusion and to understand the underlying mechanism of this association. Further research is also required to investigate the role of individual antibacterial agent is in the development of acute pancreatitis. We conducted a comprehensive review of four different globally available databases. We will describe each database, identify the advantages and disadvantages, and the type of research questions that can be answered using the database. Included

Project (NIS-HCUP)
The largest all-payer inpatient care database in the United States is said to be  5 This database has been used by many researchers to identify and analyze different national healthcare trends, including, healthcare utilization, charges, quality, access, and outcomes. The confidentiality of individuals' information is achieved by using two procedures: (I) no patient identifiers are included in the discharge abstracts or other parts of the database, and (2) researchers or users of the database agree, through data use agreement, to use the data for research and statistical purposes without any attempt to identify individuals. To enable researchers to perform a study that spans multiple years, NIS created NIS-trends supplemental files (NIS-Trends) . .Jt is recommended to include only years after 1993 to perform ~tudy with trend analysis. 12 Since NIS is a uniform, multiple-year and multi-state database, it could be used to explore different topics such as quality of care, medical treatment effectiveness, or healthcare cost (Table III. The Veterans Affairs (VA) medical database is considered from the largest medical databases in the United States. 24 -26 It contains information for more than seven million veterans and their dependents. The VA Information Resource Center (VIReC) is responsible for assisting researchers on how to use the database and giving full information about the datasets in AITC. In addition, VIReC provides detailed information in different areas such as requesting access to the datasets, programming tips, and how to connect to the Austin mainframe to run analyses. 14 , 30 All medical variables are described in detail and determined in which file these variables are available. 14

Data structure
There are different data sources within the VA database that can be utilized by researchers to conduct their studies (Table III. Any patient who has been seen in any of the VA outpatient clinics has a record. The It has three levels of aggregation: VHA DSS Production databases, VHA DSS Report summary, and VHA DSS National Data Extracts (NDE). 38 77

VHA DSS Production Databases
DSS production database contains data regarding costs; clinical information, such as resource utilization and patient outcomes; and workload captured over different levels (e.g. encounter, day and laboratory results). The software also has tools to help the researcher or the team staffs to perform a variety of analyses. The researcher can study a pattern of care for inpatient stay, following a cohort of patients with a specific outcome over time; study a specific topic related to laboratory results; . d f . ·1 h . 37 38 or a multitu e o s1m1 ar researc questions. ' The DSS Pharmacy NDEs are a major source for researchers to conduct research requiring information on medications. They are not the only datasets that include medication information, but they are the largest.

Other pharmacy resources
In addition to the DSS Pharmacy National Data Extract, other important resources include VISTA system, DSS National Pharmacy Extract "Planned" datasets, and the Pharmacy Benefits Management (PBM) group.   Also, permission is required to utilize the required datasets. A specific form (VA Form 9957) is required to identify and access the required dataset(s). These datasets are housed and maintained by the Austin Information Technology Center (AITC).

Advantages and Limitations
Veterans Affairs medical databases are a unique and comprehensive databases (Table III.13). However, the VA databases have some limitations, such as the validity of ICD-9 for some diseases. For example, based on a study done in 2009, the ICD-9 for gout disease is not accurate. 47 Another study found that using ICD-9 to identify viral hepatitis (type B) is not accurate in the VA database. 48 Nevertheless, many diseases have an accurate ICD-9, such as opportunistic infections and serious bacterial infections, chronic liver diseases, viral hepatitis (type C), HIV, and many other diseases. 48 , 49 Not all demographic data such as height and weight is available directly by using medical SAS® databases or DSS databases. The Corporate Data Warehouse (CDW), another medical database, should be used to access these variables. CDW requires researchers to submit certain documents if they require access.

The Health Improvement Network (THIN) database
In the United Kingdom (UK), there are more than fourteen databases containing clinical and demographical data for patients in the UK. 50 Most of European countries, such as the UK, Germany, the Netherlands, and Norway, have a national healthcare system. Having a national healthcare system is beneficial for a comprehensive and ideal healthcare database. In the UK, the National Health Service This software is the same software that is used by GPRD. 55 and is used by over 95% of the primary care practice and health boards in the UK. 54 Although the THIN database is a new database, it is considered a good alternative resource for medical research since it has informative, valuable, and has complete data for each patient. Considering how new this database is, it has a very high possibility of becoming one of the premier healthcare databases in the UK.

THIN Data
In UK, the general practitioner is considered as gate keeper for the patients as they are required to visit their GP if they need medical assistance. There are only a few exceptions to this rule. When the patient visits the GP, the GP will assess the patient's case and decide to either issue a prescription, refer the patient to a specialist, refer them for tests, refer to an emergency room, arrange for hospital admission, or offer advice. A patient does not need to go to the GP first if he/she is receiving services from an emergency room, a dentist, an optician, or a family planning and sexual health clinic. 56  Additionally, all studies that are going to be conducted using the THIN database need to be approved by MREC. In the near future, the approval process will be changed, whereby the scientific protocol review committee (SPRC ) will replace MREC. Unless these studies need to collect additional data or need validation, MREC's approval is still required. 53  and symptoms recorded at the GP office as well as the patient unique identifier. These 87 files include the event date, diagnosis, diagnostic data (e.g. X-ray, CT scan, MRI and others), episode type, location of consultation, medical history, and referral data (Table III.14). 53 ' 58 ' 59 If the patient is transferred to secondary care, all the data from their secondary care is sent back to the GP and entered into the system. 58 This data consists of very detailed information including type of specialty; hospital admission; hospital discharge, diagnosis, and medication; outpatient consultation diagnosis; . . d t t t t 52 ' 58 investigation; an rea men ou come.
All health conditions and diseases in THIN databases are coded using the READ clinical classification version 2. 60 READ codes are alphanumeric codes that define diseases using a hierarchical nosologic system (Table III.15). 61 - 65 The code terms used in READ are related to the observations (signs and symptoms), diagnoses, procedures, and laboratory and radiologic tests. 53 The therapy file includes information about prescription data. After the GP enters the prescription data, one copy is sent to the pharmacy and one printed copy is given to the patient to take it to the pharmacy. 58 Using this procedure creates a medication record for each patient that Databank' s Multilex® coding system and the BNF coding system (Table 111.16). 58,66,67 There are three codes for BNF (BNF code 1, BNF code2, BNF code3) each 88 representing different groups of medications. Therefore, some medications that may have more than one BNF code. For example, aspirin can be classified as a nonsteroidal anti-inflammatory drug (NSAID) and an anti-platelet. 68 The Medication that is prescribed outside the GP office by hospital physicians and specialist physicians usually is not recorded in the system, excluding it from the THIN database. However, if the patient continues to take the medication that he/she started taking in the hospital, the drug will be recorded in the database because the patient will follow up with the GP. 58 Some of the medications data from outpatient specialty care may also be found in the GP summary. 58 The additional health data (AHD) file contains supplemental information such as patient unique identifier, GP consultations, details from other healthcare interventions, patient's height and weight, allergies, vaccinations and immunization, contraceptive prescriptions, pregnancy, birth details, death details, laboratory tests and results, and life style information including smoking and alcoho 1 intake. 53 ' 59 Recently, the pathology laboratory has been electronically linked to many practices making it easier to record patient tests results and store them in their records.
Some of the data in AHD can be found in other files. For example, data on smoking and alcoho 1 use can be found in medical records, and contraception data can be found in therapy files and in medical files. 68 In addition to these main files, EPIC creates Postcode Variable Indicators (PVI). PVI files provide unidentified postcodes which link the area based on socioeconomic, ethnicity, and environmental characteristics to help researchers who are planning to conduct a study using THIN to understand the differences between the areas to avoid any potential bias or confounders that could affect the study. Under the PVI file, there are information related to the degree of deprivation, ethnicity, and the degree of pollution (with particulate matter 10 (PMI 0), sulfur dioxide (S02) and nitrogen dioxide (N02). Each one of these three variables is matched to the UK postcode. The ethnicity is categorized into white, black, mixed, Asian, and other. 53,58,59 Additional information about any patient may be obtained through the Additional Information Service (AIS), which includes· information such as death certificate, unidentified questionnaires filled out by the patient or by the GP, and a specific intervention (e.g. specific diagnostic test to confirm a specific diagnosis). 59

Advantages and limitations of THIN database
The THIN database has many advantages for both the health system in the UK and for research studies (Table III. 17). It helps to improve the quality of clinical data recording and, subsequently, will improve the quality of healthcare in the UK. THIN offers many training sessions for the users (GP and researchers) to become familiar with the database. 77 THIN is a computerized database, allowing for the compilation of complete and updated data for each patient. Therefore, any investigator can obtain new and additional patient information for a study. It is a population-based database, so selection bias will be minimized making epidemiological study more valid. 58 In addition, recall and interviewer biases are not a problem since all information that has been collected is not based on patient recall. Additionally, in conducting epidemiological study, the cases and controls are derived from the same type of 92 population. Therefore, all the criteria inside the database were applied to all patients (including cases and controls). Another advantage of the THIN database is that it contains longitudinal data allowing studies with long follow up to be conducted easily.
Another advantage of THIN database is the size of database. It has a large number of patients enabling many health conditions, including rare conditions, to be studied. The THIN database has been validated through many studies, and these studies have found the information inside THIN database to be valid and complete. 55 , 75 This is one of the important issues for any database. However, additional studies are still needed to validate different disease states.
Finally, researchers have the ability to access the original medical records pertaining to the topic being studied. The researcher can request a copy of a patient's medical record (non-electronic). There will not be any identification for the patient on the medical record, but this service can help the researcher in many aspects such as validating the information and getting detailed information about the patient. 68 However, there are some disadvantages to using the THIN database. As a UK

The Norwegian Medical Database
Norway, a Scandinavian country in Europe, has a population of 4.8 million as of 2009. 79 The population has increased by 1. 5 million since 19 5 0. 80 This increase in the growth rate was caused by a rising birth rate, prolonged life expectancy, and an increased number of immigrants to Norway. Norway is considered to be among the highest ranked nation in the number of immigrants. Norway has many strengths in its educational and economic systems. 81 More than 57% of the population aged 16 and more had completed their secondary school, placing Norway among the most educated .
Id .d 81 82 countnes wor w1 e. ' The healthcare system in Norway is structured on three levels: the national level, the regional level, and the local level. The national level is responsible for establishing and providing necessary legislation, determining the national health policy, and funding the health services. 81 ' 83 The regional and local levels are responsible for providing the health care services through a network of hospitals, primary care clinics, psychiatric clinics, pharmacies, outpatient clinics, dental clinics, ambulance services, and other health services. 81 Overall, the regional authority is responsible for specialist healthcare while the local level is responsible for primary health care. The Ministry of Health and Care Services is responsible at the national level. Norway has an excellent environment to have a good healthcare system. Its national health system grants each citizen the right to health services access regardless of their geographic location or economic and social status. 81

Patient pathway
Patients who seek medical advice or therapy must visit the general practitioner (GP) first . Approximately, 99% of the population is registered with one general practitioner. The ministry of health gives the freedom to the patient to choose a GP among from a list in their local area. 81 ' 85 The GP is usually located at the municipality' s level and acts as a gatekeeper to further medical care. The GP will take one of the following actions: diagnose the patient and give the suitable treatment or make a referral if the patient needs to be seen by a specialist physician. The specialist physicians are located at the specialized healthcare centers or institutes. The specialized healthcare centers include hospitals and specialized medical services such as laboratory services, radiology services, and special care for alcoholics and or persons with drug addiction. Furthermore, it includes somatic and psychiatry 96 institutes. 81 ' 84 ' 86 Each regional healthcare authority is responsible for running these centers to deliver high quality specialized healthcare to each patient in the region. 81

Registries and databases
As mentioned previously, Norway has a good healthcare system to build good  (Table III. Previous databases were used to establish a new monitoring system that combined many health indicators that make it a comprehensive database (Table III. 19). 92 Since the official registries and databases in Norway have the patient unique personal identification number, it became a very useful tool to link these registries together to make this monitoring health system. Additionally, this method assists linking the health information with other information such as education and socioeconomic data. It is called Norhealth, or the Norwegian Health Information System, and it is used to monitor health related conditions over time. Having a system like Norhealth helps to produce a very good health monitoring system and high quality research using the variety health indicators from different types of registries. In October 2003, the final regulation on the Norwegian prescription database was released to achieve many objectives. The regulation obligates pharmacies in Norway to submit all prescription data "electronically" to NIPH on a monthly basis. 95 The main objective ofNorPD is to collect and process the data on drug utilization for both humans and animals. In addition to this main objective, NorPD has several aims (Table III.

Advantages and limitations
The health system in Norway has many factors that make it one of the most unique health systems worldwide. These factors help to establish and maintain distinctive databases and registries. It is well known that having a national "social" health system is considered an important factor to have an ideal health database. 2 ' 3 Norway has several advantages pertaining to the databases. It has several databases that specialized in areas such as cancer, medical birth, death, and communicable and non-communicable disease. Having a specialized database in each area, it is important to have a patient's complete and comprehensive data within these specialized databases. For instance, the data in the cancer registry is almost 100% complete. 101 In addition to having specialized databases, Norway has more than eight databases that cover most health areas in addition to medication data. Since most individuals are registered within one primary care clinic, most patients have their health information in these databases. Therefore, any study done using these databases will be considered externally valid (i.e. can be generalized to the Norwegian population) All these databases have the eleven-digit p€rsonal identification numbers (person nummer) that makes it easy to link all these databases together simply by using this number.
Therefore, it is possible to conduct different epidemiological researches in different areas. In addition, performing a system of monitoring allows researchers to follow the trend of the diseases in Norway, and post-marketing surveillances to detect the signals for adverse drug reactions as well. Furthermore, it is a relatively inexpensive database for researchers to access.

Conclusion
In conclusion, each database has its own strengths and limitations.NIS can answer many research questions related to healthcare utilization and healthcare cost, but it is not the best database to answer questions related to medication exposure since there is no detailed information about medications. The latter research question can be investigated thoroughly and accurately by using the VA databases because these databases contain enough data on patient medications. Many other research questions can also be answered using VA databases because they have ample data on most clinical aspects as well as economic aspects. Pharmacoepidemiologic, pharmacoeconomic, and other behavioral studies can be performed using the VA databases. However, any study conducted using VA database will only be generalized to the VA population. This problem is not a big issue in countries with a social healthcare system such as the UK and Norway. The conclusions of studies that use the THIN database in the UK can be applied to the entire UK population. This is also true of studies that used the Norwegian database. Th~ THIN database shares similar characteristics of the VA database in its ability to answer many research questions related to diseases and medications. Finally, the Norwegian databases are very interesting and well-organized databases. The major strength of these databases is that they can be linked together to produce large amounts of patient information to conduct many studies. Researchers from inside as well as outside Norway should start to utilize these databases. The major difference between U.S databases and European databases is a result of the type of healthcare system in each country. As mentioned before, the social healthcare system is an ideal environment to produce a good database. This can be found with the THIN database and the Norwegian databases.
The VA database also has similar attribute because VA health care system somewhat resembles a government-run, universal healthcare system.       2009 They were implemented in 1999 and it has several variables such as number of X-rays, discharge DRG, different type of costs and other variables.

Outpatient
Outpatient datasets include services that are OutPat FY1999not available in the NPCD.
2009 They have more detailed cost data and some clinical data. The files are divided to different group because of the big size.

Treating
Treating specialty datasets could have more TRT FY1999-Specialty than one record for each hospital stay. 2009 They are divided into different sections based on the month and treating specialty of the provider for each stay. It is associated with the location such as medical care, surgical ward or long-term care unit. Laboratory Laboratory results dataset includes LAR FY2000results laboratory results for defined number of 2009 laboratory tests. It has both inpatient and outpatients encounters.

115
Labo rat Laboratory datasets include record for every LAB FY2002ory individual "patient" laboratory tests for both 2009 inpatient and outpatient encounter during the extract period.

Radiology
Radiology datasets include data such as RAD FY2002type, cost and number ofradiological 2009 procedures that performed in the Diagnostic Radiology and Nuclear Medicine Departments for every encounter.

Pharmacy
Pharmacy datasets provide data on drugs PHA FY2002that used in both inpatient and outpatient.
2009 They include prescription, unit dose, and IV pharmacy detail.   Creatinine Clearance (Ml/min) FY2000 11 Digoxin (Ng/ml) FY2000 7 AST "Aspartate Transaminase" (U/L) FY2000 9 ALT "Transferase Alanine Amino" (U/L) FY2001 45 Hematocrit (%) FY2002 50 INR "International Normalized Ratio" FY2003 52 Creatinine GFR (ML/MIN/1. 73M2) FY2006 66      To explain the pattern of medication utilization and monitor the trends of utilization over time To function as a base resource for drug related statistics to be used by authorities to assess the quality of drug prescription, controlling and planning To function as an internal control for prescribers to improve the quality of prescribing practice.
To function as the main resource for medication and pharmacoepidemiological research.