New Results

Defining Major Depressive Disorder Cohorts Using the EHR: Multiple Phenotypes Based on ICD-9 Codes and Medication Orders

Wendy Marie Ingram, Anna M. Baker, Christopher R. Bauer, Jason P. Brown, Fernando S. Goes, Sharon Larson, Peter P. Zandi
doi: https://doi.org/10.1101/227561

ABSTRACT

Objective Major Depressive Disorder (MDD) is one of the most common mental illnesses and a leading cause of disability worldwide. Electronic Health Records (EHR) allow researchers to conduct unprecedented large-scale observational studies investigating MDD, its disease development and its interaction with other health outcomes. While there exist methods to classify patients as clear cases or controls given specific data requirements, there are presently no simple, generalizable, and validated methods to classify an entire patient population into varying groups of depression likelihood and severity.

Materials and Methods We propose an electronic phenotype algorithm that classifies patients into one of five mutually exclusive, ordinal groups, varying in depression phenotype. Using data from an integrated health system on 278,026 patients from a 10-year study period we demonstrate the convergent validity of these phenotype constructs by presenting multiple lines of evidence associated with depression.

Results Convergent validity is derived from expected patterns in health care utilization, psychiatric prescriptions, indicators of suicidality, diagnoses of serious comorbidity, mortality, symptom severity, and finally, polygenic risk scores.

Discussion The algorithm is generalizable to most EHR data sets because it requires only International Classification of Diseases (ICD) diagnostic codes and medication orders and can be used for stratification of an entire patient population.

Conclusion Careful consideration must be given to the definitions of patient cohorts when utilizing EHR data, particularly when classifying subjects with heterogenous disorders such as MDD. This algorithm may prove useful to others that wish to study depression in entire patient populations with EHR data.

BACKGROUND

Depression is a highly prevalent mental illness that accounts for $43 billion in medical costs annually and is a leading cause of disability[1]. Depression has been linked to worse outcomes and increased healthcare utilization for numerous common medical disorders[2–6]. However, depression is a heterogeneous disorder, and its etiologies remain poorly understood [7]. There is an urgent need to better understand the causes and course of depression in order to develop more effective treatment and prevention strategies. Electronic Health Records (EHR) from large integrated health systems now offer the opportunity for researchers to conduct unprecedented, large scale studies of patients in real-world settings[8–13]. Critical to these pursuits are phenotypic algorithms that correctly distinguish who has the disorder within the patient population[14, 15]. Depression is a particularly difficult phenotype to define and studies often use heterogeneous criteria when utilizing EHR data to identify patients with depression[16–21]. As with any phenotypic algorithm, the challenge is to validly define depression with high sensitivity and specificity, limiting both false positive and false negative classification of patients [22]. An additional complexity is the fact that depression may not be a binary phenomenon, but rather it may exist on a continuum with a range of severity in a population.

There are at least four potential sources of information from the EHR for defining depression: a) International Classification of Diseases (ICD) diagnosis codes; b) depression screening measures; c) medication orders; and d) clinical notes. While some studies use only ICD diagnosis codes to identify patients[21, 23], others have demonstrated that using these alone has inferior sensitivity and precision when compared with combinatorial models that use multiple sources of information [24]. National recommendations that adults be screened for depression annually has increased the availability and use of symptom questionnaires[16–18, 25] such as the Patient Health Questionnaire (PHQ-9)[26]. However, implementation of such screening measures has been fairly recent, is not standardized, and shows limited agreement with ICD codes for depression [19, 20]. Phenotyping algorithms may also use information from medication treatment codes [27–29]. Unfortunately, there may be a long delay between when patients with depression first experience the onset of symptoms and ultimately receive care, including with medication[30–33]. Additionally, antidepressants may be prescribed for a variety of comorbid mental [34] and non-mental health indications[35], such as tobacco use cessation[36] or chronic pain[37], which complicates its use for reliably identifying depression [38]. Lastly, the use of natural language processing (NLP) on clinical notes has great promise for classifying psychiatric disorders[24, 39–42]. However, the generalizability of these methods may be limited due to data sets that do not have the number or types of notes required or contain only deidentified data. Finally, most methods involve the exclusion of a sizeable number of patients with uncertain status, which prevents clinically relevant population-wide studies that include and classify all patients in the population.

OBJECTIVE

This study examines a novel phenotyping algorithm for defining depression along a continuum using EHR data and evaluates their construct validity with other indicators of health that should correlate with depression. We focus on definitions based on ICD diagnosis codes and medication order data because these sources of data are more readily available than depression screening data and/or clinical notes and thus the resulting definitions are more widely generalizable.

MATERIALS AND METHODS

Study data and analysis

This study included de-identified Electronic Health Records (EHR) data obtained from January 1st, 2005 to September 30th, 2015 (10.75 years) for patients seen in the Geisinger Health System, an integrated health care system located in central Pennsylvania. The Geisinger system has a stable patient population whose EHRs have been collected in a central data warehouse and are available for clinical and research purposes [43–46]. The end date of the study period was chosen based on the transition from ICD-9 codes to ICD-10 codes in this hospital system. Adult patients 18 years or older at the beginning of the study, 90 years or younger at the end of the study, who had a Geisinger Primary Care Physician (PCP) at any point during the study period, and had at least one outpatient visit within the system during the study period were included in the study cohort (n=278,026) Demographic information, medication order histories, and details of all outpatient, Emergency Department (ED), and inpatient encounters were obtained on these patients. The study was approved by the Geisinger Internal Review Board as non-human subjects research.

We used domain knowledge of depression clinical care to implement an algorithm for partitioning all patients into one of five ordinal phenotype groups reflecting decreasing/increasing likelihood and/or severity of depression based on the available ICD-9 diagnosis codes and medication orders, Then, to evaluate the convergent/divergent validity of these phenotype groups, we examined how they related to other health care related characteristics thought to be associated with depression. These other characteristics included measures of health care utilization, medical history, depressive symptoms, and polygenic risk for depression. All analyses and visualizations described in further detail below were conducted in R (2017, R Core Team, Vienna, Austria) and GraphPad Prism 6 (La Jolla, CA).

Depression phenotype algorithm

To implement the phenotype algorithm (Figure 1), patients were first grouped into “Major Depressive Disorder” (MDD) if they received a diagnosis of Major Depressive Disorder (ICD-9 codes: 296.20, 296.21, 296.22, 296.23, 296.24, 296.25, 296.26, 296.30, 296.31, 296.32, 296.33, 296.34, 296.35, 296.36, 296.82) one or more times as an ED or inpatient discharge diagnosis or on their problem list, or two or more times within a 2-year period as an outpatient discharge diagnosis. Second, the remaining patients who did not meet previous criteria were then grouped into “Other Depression” (OthDep) if they received a diagnosis of a depressive disorder not elsewhere classified (ICD9 code: 311) one or more times as an ED or inpatient discharge diagnosis or on their problem list, or two or more times within a 2-year period as an outpatient discharge diagnosis. Third, the remaining patients who did not meet any of these previous criteria were then grouped into “Multiple Antidepressants No Depression” (MultiRx) if they received two or more antidepressant medication orders (RxNorm classification “Antidepressant”) during the study period that were not prescribed for common indicatons unelated to depression, including tobacco use disorder (305.1) and hereditary and idiopathic neuropathy (356). Fourth, the remaining patients who did not meet any of these previous criteria were then grouped into “Miscellaneous Antidepressants NoDepression” (MiscRx) if they received any other antidepressant medication orders not captured in the previous grouping. Finally, all other remaining patients were grouped into “No Depression” (NoDep).

Fig. 1
Fig. 1

Methodological design with five mutually exclusive constructs including multiple depression phenotypes and one control group with no evidence of depression. “according to rules” refer to details of the Depression Phenotype Algorithm described in the Methods section. “311 codes” refer to the ICD-9 code for Depressive disorder, not elsewhere classified. “TUD” stands for tobacco use disorder.

Healthcare utilization

Depression is associated with increased healthcare utilization[47–49]. As a measure of healthcare utilization, we calculated the average number of visits annually across different health care settings and compared these across the groups. To calculate the average number of visits in each healthcare setting per patient per year, we tallied the number of total visits in the outpatient, ED, non-psychiatric inpatient, and psychiatric inpatient settings during the entire study period for each group and divided this by the total number of patients in that group and the length of the study period (10.75 years), resulting in the average yearly visit frequency.

Medical features

We then evaluated several medical features that are known to be associated with depression, including medications, comorbidities and mortality. Depressed patients are often prescribed antipsychotics to augment their antidepressant medication[50], and antianxiety agents to treat anxiety, which is a well-known comorbid condition with depression[51]. We determined the percent of patients in each group that had ever received at least one order for these medications as classified in RxNorm (“Antipsychotic” and “Antianxiety Agent”, respectively). Depressed patients are known to have higher rates of suicidality[52, 53] and mortality[54, 55] than non-depressed patients. We used the discharge diagnosis codes from encounter records to determine the percent of patients with ICD-9 codes for suicide and self-inflicted injury (E950 - E958). We used the date of death to determine the percent mortality of each group during the study period. Substance abuse was captured using discharge diagnosis codes from encounter records to determine the percent of patients in each group that had ICD-9 codes for alcohol and drug abuse or dependence (303, 304, and 305, excluding tobacco use disorder 305.1). Finally, as depressed patients have greater overall burden of medical comorbidities than non-depressed patients[56], we calculated the Charlson Comorbidity Index (CCI) score for each patient. The CCI contains 19 categories of serious comorbidities that predicts the 10-year mortality of patients[57].

Depression symptoms

In 2012, Geisinger Health System began implementing universal screening for depression with the Patient Health Questionnaire 2 (PHQ-2). If patients endorse either of the two screener questions, they are then asked the additional 7 questions (PHQ-9). The PHQ-9 is a validated instrument for assessing current depression symptom severity and has a tiered rating scale based on total score (0: no depression, 1-9: mild, 10-14: moderate, 15-19: moderately severe, 20+: severe)[58, 59]. For all patients that had one or more PHQ-2/9result in their EHR (n=170,618; 61.4%), we identified their maximum score and determined the percent of patients in each group with maximum scores in the different PHQ-2/9 scoring categories.

Polygenic risk scores

Geisinger Health System has recruited over 90,000 patients to participate in a genetics study called MyCode[43]. A subset of the study population described above have participated in this study, allowing us to calculate polygenic risk scores (PRS) for comparison across depression groups. Using publicly available summary results from a recently published genome-wide association study (GWAS) of MDD [60], we calculated PRS for each patient in our data set that had genetic data available (n=52,775; 19%). Polygenic risk scores for MDD were calculated using PRSice-2[61] at eight predetermined p-value thresholds: 1, 0.5, 0.1, 0.05, 0.01, 0.001, 0.0001, 0.000001. As results were similar across all p-value thresholds (Supplementary Table 1), we only report results for the nominal threshold of p=0.05.

RESULTS

We developed and implemented an algorithm for grouping patients based on ICD9 codes and medication orders into one of five ordinal groups, varying in likelihood and severity of depression, which we named from most to least likely/severe: “MDD”, “OthDep”, “MultiRx”, “MiscRx”, and “NoDep” (Figure 1). Of the total patient population, each group accounts for 3.2%, 17.4%, 11.1%, 10.8%, and 57.5%, respectively. Summary statistics on sex, race, marital status, age at beginning of the study and length between first and last encounters are shown for the total patient population and all five groups in Table 1. The sample is 54.0% female, 95.7% white, 60.9% married, median age 45, and median length between first and last encounter is just over 7 years. As expected, depression occurs more commonly in females, reflected by an increased Female to Male ratio in each “affected” group. Patients in the MDD and OthDep groups are less likely to be “Married” compared with those in the NoDep, MultiRx, and MiscRx groups. Age at the beginning of the study does not differ substantially between groups (median: 45, third quartile: 57 years), although it is slightly less for those in the MDD group (median: 44, third quartile: 55 years). The length between first and last encounter varied across groups, with a much higher percentage of patients in one of the four “affected” groups having at least 7 years of observation than those in the NoDep group. It is also of note that the vast majority of patients in the MDD and OthDep groups received at least one antidepressant medication order, 97.2% and 93.0%, respectively.

View this table:
Table 1.

Characteristics of the total sample and five depression phenotype groups

Healthcare utilization

The average number of outpatient visits/year decreased monotonically across the five phenotype groups from a high of over 5.5 visits/year for MDD patients to a low of fewer than 2 visits/year for NoDep patients (Fig 2A). The pattern was similar albeit with lower overall averages for ED visits (ranging from 0.24 to 0.06 visits/year, respectively) and non-psychiatric inpatient visits (ranging from 0.11 to 0.03 visits/year, respectively) (Figures 2B and 2C). The most striking difference across the five phenotype groups was observed for psychiatric inpatient visits (Figure 2D). MDD patients had an average of 0.03 such visits/year, which was 10 times greater than that for OthDep patients, the next highest group, while the rates were negligible for the remaining MultiRx, MiscRx and NoDep groups.

Fig. 2
Fig. 2

Healthcare utilization patterns for the five depression phenotype groups. (A) Outpatient visit frequency; (B) Emergency Department (ED) visit frequency; (C) Non-psychiatric inpatient visit frequency; (D) Psychiatric inpatient frequency.

Medication orders and comorbidities

Similar to the utilization patterns, the percent of patients that received antipsychotic (Figure 3A) or antianxiety agents (Figure 3B), decreased monotonically across the five groups from a high for MDD to a low for NoDep. Approximately 44% of MDD compared to 5% of NoDep patients were prescribed antipsychotic medications, while 31% of MDD compared to 5% of NoDep patients were prescribed antianxiety medications. A similar pattern was observed across the five phenotype groups for the percent of patients with substance use and abuse (Figure 3C) and mean Charlson Comorbidity Index scores (Figure 3D). The pattern differed, however, for suicide related diagnosis codes and overall mortality. Almost all suicide related diagnosis codes were noted in MDD and OthDep patients, while the percent of patients with such codes were negligible in the MultiRx, MiscRx, and NoDep groups (Figure 3E). As for overall mortality, interestingly, the highest rates were observed for patients in the OthDep group. (Figure 3F).

Fig. 3
Fig. 3

Medical features across the five depression phenotype groups. (A) Antipsychotic and (B) antianxiety medication prescriptions, (C) substance use disorder or dependence codes, (D) mean Charlson Comorbidity Index scores, (E) suicide codes, and (F) mortality.

Medication orders and comorbidities

Similar to the utilization patterns, the percent of patients that received antipsychotic (Figure 3A) or antianxiety agents (Figure 3B), decreased monotonically across the five groups from a high for MDD to a low for NoDep. Approximately 44% of MDD compared to 5% of NoDep patients were prescribed antipsychotic medications, while 31% of MDD compared to 5% of NoDep patients were prescribed antianxiety medications. A similar pattern was observed across the five phenotype groups for the percent of patients with substance use and abuse (Figure 3C) and mean Charlson Comorbidity Index scores (Figure 3D). The pattern differed, however, for suicide related diagnosis codes and overall mortality. Almost all suicide related diagnosis codes were noted in MDD and OthDep patients, while the percent of patients with such codes were negligible in the MultiRx, MiscRx, and NoDep groups (Figure 3E). As for overall mortality, interestingly, the highest rates were observed for patients in the OthDep group. (Figure 3F).

Depression symptoms

A subset of the study population was screened at least once with the Patient Health Questionnaire (PHQ), a well validated depression measure (Figure 4). Patients in the MDD group were the most likely to have been screened (73.4%, n=6421), followed by those in the OthDep (68.7%; n=33,229), MultiRx (69.1%; n=21,419), MiscRx (62.9%; n=18,865), and finally NoDep (56.7%; n=90,684). The majority of those in the NoDep group scored a maximum of 0, indicating no depression, and only 2.6% scored 10 or higher, indicating moderate, or more severe depression. Those in the MiscRx and MultiRx groups were remarkably similar to one another at all possible scores. This contrasts with the more severely “affected” groups MDD and OthDep. Over a third of the MDD group and nearly a quarter of the OthDep group had a maximum score of 10 or higher.

Fig. 4
Fig. 4

Depression symptoms across the five depression phenotype groups. Percentage of patients per group whose maximum score on the Patient Health Questionnaire 2 or 9 (PHQ-9/2) fell in the categories shown.

Polygenic risk scores

As depression has a prominent heritable component [62], we used polygenic risk scores (PRS) derived from an external GWAS [60] to further validate our phenotype groups for those patients in this study who had available genome-wide genotype data. As shown in Figure 5, we found a gradient of increased PRSs across the five phenotype groups, with the highest PRSs seen in the MDD group and the greatest difference seen between the MDD and NoDep groups (P < 2.2 -16, R2 =0.8%).

Fig. 5
Fig. 5

Scaled Polygenic Risk Scores for Major Depressive Disorder correlate with and distinguish between depression constructs when compared with NoDep group. P-value threshold 0.05, includes 20,962 SNPs. Comparing NoDep to both MDD groups combined results in R2=0.008. Welch two sample t-test between each group and NoDep results in statistically significantly differences in each comparison. Median values displayed in bold at center of boxplots overlaid on jittered dot plots.

DISCUSSION

Here we present a novel electronic phenotype algorithm using ICD codes and medication orders in EHR data, resulting in five mutually exclusive and ordinal groups for categorizing patients with depression across an entire population. These phenotype groups demonstrate convergent validity as assessed by significant differences between them in a range of clinically relevant characteristics that are expected to correlate with depression [63, 64], including healthcare utilization [47], medical features (such as treatments [65], comorbidities [66, 67], and mortality), depression screening results, and polygenic risk scores. Interestingly, these clinically relevant characteristics tended to vary in a “dose-response” like fashion across the phenotype groups consistent with the fact that the groups defined by increasingly more stringent criteria identified patients with increasing likelihood and/or severity of depression. These findings demonstrate that it is possible to use diagnostic codes and medications orders in EHR data to validly categorize patients with respect to depression across the entire population.

With the wide spread adoption of EHRs, there is growing interest in identifying patients with depression for subsequent research using data available from the EHRs. As a result, a number of previous studies have attempted to do this employing a variety of methods that use in isolation or combination ICD9 diagnosis codes, antidepressant prescription orders, and NLP of progress notes. Many of these studies have not validated their algorithms [18–21, 23]. There are a few studies, however, that have attempted to validate their algorithms by comparing identified cases and controls against a putative gold-standard diagnosis established by expert review of 50 or 100 medical charts per group [17, 24, 39]. Our approach differs from these previous efforts in that we sought to establish convergent rather than criterion validation of our algorithm, which allowed us to characterize and compare the full range of depression in the entire sample rather than a small sub-sample of putative cases and controls which presumably represent the extremes of the phenotype in the population. In addition, other than one study which used genetic data to try and validate an algorithm for defining a bipolar disorder phenotype [40], our study is the only one as far as we know to use genetic data to validate an algorithm for defining depression.

This study has several limitations that warrant discussion. First, with regard to our examination of healthcare utilization patterns and medical features, we assumed that patients contributed follow-up over the entire study period. We did this to simplify our estimates, though it is possible that a number of patients entered or left the Geisinger service area during the study period. Second, with regard to our algorithm, we categorized patients based on the most severe criteria met over the study period and assumed they belonged to that group for the entire period. This ever/never approach does not allow for change over time and does not take full advantage of this rich longitudinal data, which may be interesting to study using longitudinal data analysis methods in the future. Our algorithm was also based on ICD-9 codes, as opposed to ICD-10 codes which are the current version in use for diagnosis and healthcare billing. Fortunately there are many resources available for mapping between ICD-9 and ICD-10 codes that can be leveraged to create an equivalent list of ICD-10 codes to the list and codes used here [68, 69]. Third, with regard to the sample under study, it consisted of predominantly white patients from rural communities in central Pennsylvania. As a result, there may be questions about the generalizability of our approach to other more diverse and urban populations. In addition, there are other healthcare systems in the service area and thus the data we used to categorize patients may not have completely captured all medical utilization. Claims data may provide an additional source of information about healthcare utilization that would be useful in future studies.

This study also benefits from several significant strengths. One of the major strengths of this study is the large longitudinal data from a stable patient population in an integrated health care system. With extended and extensive data available from multiple health care settings (primary, emergency, and hospital care), we were able to validate our simple phenotype algorithm by comparison with a range of different measures. Additionally, our algorithm allows for classification of all patients in a given population across a spectrum of depression severity, as opposed to other approaches which often seek to classify only the extremes of the population to the exclusion of patients with uncertain binary case versus control status [39]. Finally, our algorithm uses only diagnosis codes and medication data to validly classify patients with regard to depression and therefore may be more generalizable to other systems that don’t have ready access to data on screening measures or from natural language processing of clinician notes.

CONCLUSION

The electronic phenotype algorithm presented here provides a simple and valid model for defining patients with varying severities of depression and may be useful for researchers who wish to examine the effects of depression in entire patient populations. The five mutually exclusive, ordinal groups demonstrate differences that are expected to correlate with depression. We found correlations with utilization patterns, treatments, comorbidities, mortality, depression screening instruments for symptom severity, and polygenic risk scores. These constructs are generalizable to any data set that has both diagnosis codes and medication orders and at least two years of data. Ultimately, the definition of depression phenotypes will depend upon the goal of the research, and we present one possible method that demonstrates convergent validity, generalizability, and inclusivity.

ACKNOWLEDGEMENTS, COMPETING INTERESTS, FUNDING

Funding

Author WMI is funded by the National Institutes of Health (T32MH014592-41 Psychiatric Epidemiology Training Program)

Conflict of Interest

The authors declare that they have no conflict of interest.

REFERENCES

  1. 1.↵
    Maurer DM (2012) Screening for Depression. Am Fam Physician 85:139–144
    OpenUrlPubMed
  2. 2.↵
    Thomas M, Price D (2008) Impact of comorbidities on asthma. Expert Rev Clin Immunol 4:731–742. doi: 10.1586/1744666X.4.6.731
    OpenUrlCrossRefPubMed
  3. 3.
    Celano CM, Huffman JC (2011) Depression and Cardiac Disease. Cardiol Rev 19:130–142. doi: 10.1097/CRD.0b013e31820e8106
    OpenUrlCrossRefPubMed
  4. 4.
    Ang DC, Choi H, Kroenke K, Wolfe F (2005) Comorbid depression is an independent risk factor for mortality in patients with rheumatoid arthritis. J Rheumatol 32:1013–1019
    OpenUrlAbstract/FREE Full Text
  5. 5.
    Young BA, Von Korff M, Heckbert SR, Ludman EJ, Rutter C, Lin EHB, Ciechanowski PS, Oliver M, Williams L, Himmelfarb J, Katon WJ (2010) Association of major depression and mortality in Stage 5 diabetic chronic kidney disease. Gen Hosp Psychiatry 32:119–124. doi: 10.1016/j.genhosppsych.2009.11.018
    OpenUrlCrossRefPubMedWeb of Science
  6. 6.↵
    Pouwer F, Nefs G, Nouwen A (2013) Adverse Effects of Depression on Glycemic Control and Health Outcomes in People with Diabetes. Endocrinol Metab Clin North Am 42:529–544. doi: 10.1016/j.ecl.2013.05.002
    OpenUrlCrossRefPubMed
  7. 7.↵
    Kalia M (2005) Neurobiological basis of depression: An update. Metabolism 54:24–27. doi: 10.1016/j.metabol.2005.01.009
    OpenUrlCrossRefPubMedWeb of Science
  8. 8.↵
    Jin H, Wu S, Vidyanti I, Di Capua P, Wu B (2015) Predicting Depression among Patients with Diabetes Using Longitudinal Data. A Multilevel Regression Model. Methods Inf Med 54:553–559. doi: 10.3414/ME14-02-0009
    OpenUrlCrossRefPubMed
  9. 9.
    Fiest KM, Jette N, Quan H, St Germaine-Smith C, Metcalfe A, Patten SB, Beck CA (2014) Systematic review and assessment of validated case definitions for depression in administrative data. BMC Psychiatry 14:1–11. doi: 10.1186/s12888-014-0289-5
    OpenUrlCrossRef
  10. 10.
    Simon GE, Rossom RC, Beck A, Waitzfelder BE, Coleman KJ, Stewart C, Operskalski B, Penfold RB, Shortreed SM (2015) Antidepressants Are Not Overprescribed for Mild Depression. J Clin Psychiatry 8:1627–1632. doi: 10.4088/JCP.14m09162
    OpenUrlCrossRef
  11. 11.
    Morley KI, Wallace J, Denaxas SC, Hunter RJ, Patel RS, Perel P, Shah AD, Timmis AD, Schilling RJ, Hemingway H (2014) Defining Disease Phenotypes Using National Linked Electronic Health Records: A Case Study of Atrial Fibrillation. PLoS One 9:. doi: 10.1371/journal.pone.0110900
    OpenUrlCrossRefPubMed
  12. 12.
    Affolter K, Gligorich K, Samadder NJ, Samowitz WS, Curtin K (2017) Feasibility of Large-Scale Identification of Sessile Serrated Polyp Patients Using Electronic Records: A Utah Study. Dig Dis Sci 6:1455–1463. doi: 10.1007/s10620-017-4543-9
    OpenUrlCrossRef
  13. 13.↵
    Newton-Dame R, McVeigh KH, Schreibstein L, Perlman S, Lurie-Moroni E, Jacobson L, Greene C, Snell E, Thorpe LE (2016) Design of the New York City Macroscope: Innovations in Population Health Surveillance Using Electronic Health Records. eGEMs 4:. doi: 10.13063/2327-9214.1265
    OpenUrlCrossRef
  14. 14.↵
    Sallakh MA Al, Vasileiou E, Rodgers SE, Lyons RA, Sheikh A, Davies GA (2017) Defining asthma and assessing asthma outcomes using electronic health record data: a systematic scoping review. Eur Respir J 49:1700204. doi: 10.1183/13993003.00204-2017
    OpenUrlAbstract/FREE Full Text
  15. 15.↵
    McDonald HI, Shaw C, Thomas SL, Mansfield KE, Tomlinson LA, Nitsch D (2016) Methodological challenges when carrying out research on CKD and AKI using routine electronic health records. Kidney Int 90:943–949. doi: 10.1016/j.kint.2016.04.010
    OpenUrlCrossRef
  16. 16.↵
    Pathak J, Simon G, Li D, Biernacka JM, Jenkins GJ, Chute CG, Hall-Flavin DK, Weinshilboum RM (2014) Detecting Associations between Major Depressive Disorder Treatment and Essential Hypertension using Electronic Health Records. AMIA Jt Summits Transl Sci proceedings AMIA Jt Summits Transl Sci 2014:91–96
    OpenUrl
  17. 17.↵
    Huang SH, LePendu P, Iyer S V, Tai-Seale M, Carrell D, Shah NH (2014) Toward personalizing treatment for depression: predicting diagnosis and severity. J Am Med Inform Assoc 21:1–7. doi: 10.1136/amiajnl-2014-002733
    OpenUrlCrossRefPubMed
  18. 18.↵
    Anderson HD, Pace WD, Brandt E, Nielsen RD, Allen RR, Libby AM, West DR, Valuck RJ (2015) Monitoring suicidal patients in primary care using electronic health records. J Am Board Fam Med 28:65–71. doi: 10.3122/jabfm.2015.01.140181
    OpenUrlAbstract/FREE Full Text
  19. 19.↵
    Kobus AM, Harman JS, Do HD, Garvin RD (2013) Challenges to depression care documentation in an EHR. Fam Med 45:268–71
    OpenUrl
  20. 20.↵
    Madden JM, Lakoma MD, Rusinak D, Lu CY, Soumerai SB (2016) Missing clinical and behavioral health data in a large electronic health record (EHR) system. J Am Med Informatics Assoc 23:1143–1149. doi: 10.1093/jamia/ocw021
    OpenUrlCrossRefPubMed
  21. 21.↵
    Mayer MA, Gutierrez-Sacristan A, Leis A, De La Peña S, Sanz F, Furlong LI (2017) Using Electronic Health Records to Assess Depression and Cancer Comorbidities. Stud Health Technol Inform 235:236–240. doi: 10.3233/978-1-61499-753-5-236
    OpenUrlCrossRef
  22. 22.↵
    Cole AM, Stephens KA, Keppel GA, Estiri H, Baldwin L-M (2016) Extracting Electronic Health Record Data in a Practice-Based Research Network: Processes to Support Translational Research across Diverse Practice Organizations. EGEMS (Washington, DC) 4:1206. doi: 10.13063/2327-9214.1206
    OpenUrlCrossRef
  23. 23.↵
    Gill JM, Chen YX, Lieberman MI (2008) Management of depression in ambulatory care for patients with medical co-morbidities: a study from a national Electronic Health Record (EHR) network. Int J Psychiatry Med 38:203–15. doi: 10.2190/PM.38.2.g
    OpenUrlCrossRefPubMed
  24. 24.↵
    Perlis RH, Iosifescu D V., Castro VM, Murphy SN, Gainer VS, Minnier J, Cai T, Goryachev S, Zeng Q, Gallagher PJ, Fava M, Weilburg JB, Churchill SE, Kohane IS, Smoller JW (2012) Using electronic medical records to enable large-scale studies in psychiatry: treatment resistant depression as a model. Psychol Med 42:41–50. doi: 10.1017/S0033291711000997
    OpenUrlCrossRefPubMed
  25. 25.↵
    U.S. Preventive Services Task Force (2009) Screening for depression in adults: U.S. preventive services task force recommendation statement. Ann Intern Med 151:784–92. doi: 10.7326/0003-4819-151-11-200912010-00006
    OpenUrlCrossRefPubMedWeb of Science
  26. 26.↵
    Kroenke K, Spitzer RL, Williams JBW (2001) The PHQ-9. J Gen Intern Med 16:606–613. doi: 10.1046/j.1525-1497.2001.016009606.x
    OpenUrlCrossRefPubMedWeb of Science
  27. 27.↵
    Newton KM, Peissig PL, Kho AN, Bielinski SJ, Berg RL, Choudhary V, Basford M, Chute CG, Kullo IJ, Li R, Pacheco JA, Rasmussen L V., Spangler L, Denny JC (2013) Validation of electronic medical record-based phenotyping algorithms: Results and lessons learned from the eMERGE network. J Am Med Informatics Assoc 20:147–154. doi: 10.1136/amiajnl-2012-000896
    OpenUrlCrossRef
  28. 28.
    Nadkarni GN, Gottesman O, Linneman JG, Chase H, Berg RL, Farouk S, Nadukuru R, Lotay V, Ellis S, Hripcsak G, Peissig P, Weng C, Bottinger EP (2014) Development and validation of an electronic phenotyping algorithm for chronic kidney disease. AMIA Annu Symp Proc 2014:907–16
    OpenUrl
  29. 29.↵
    Conway M, Berg RL, Carrell D, Denny JC, Kho AN, Kullo IJ, Linneman JG, Pacheco JA, Peissig P, Rasmussen L, Weston N, Chute CG, Pathak J (2011) Analyzing the heterogeneity and complexity of Electronic Health Record oriented phenotyping algorithms. AMIA Annu Symp Proc 2011:274–283. doi: PMC3243189
    OpenUrlCrossRefPubMed
  30. 30.↵
    Thompson A, Hunt C, Issakidis C (2004) Why wait? Reasons for delay and prompts to seek help for mental health problems in an Australian clinical sample. Soc Psychiatry Psychiatr Epidemiol 39:810–7
    OpenUrlCrossRefPubMedWeb of Science
  31. 31.
    Green AC, Hunt C, Stain HJ (2012) The delay between symptom onset and seeking professional treatment for anxiety and depressive disorders in a rural Australian sample. Soc Psychiatry Psychiatr Epidemiol 47:1475–1487. doi: 10.1007/s00127-011-0453-x
    OpenUrlCrossRefPubMed
  32. 32.
    Dockery L, Jeffery D, Schauman O, Williams P, Farrelly S, Bonnington O, Gabbidon J, Lassman F, Szmukler G, Thornicroft G, Clement S (2015) Stigma- and non-stigma-related treatment barriers to mental healthcare reported by service users and caregivers. Psychiatry Res 228:612–619. doi: 10.1016/j.psychres.2015.05.044
    OpenUrlCrossRef
  33. 33.↵
    Wang PS, Berglund PA, Olfson M, Pincus HA, Wells KB, Kessler RC (2005) Failure and delay in initial treatment contact after first onset of mental disorders in the National Comorbidity Survey Replication. Arch Gen Psychiatry 62:603–613. doi: 10.1001/archpsyc.62.6.603
    OpenUrlCrossRefPubMedWeb of Science
  34. 34.↵
    Shah AA, Han JY (2015) Anxiety. Continuum (Minneap Minn) 21:772–82. doi: 10.1212/01.CON.0000466665.12779.dc
    OpenUrlCrossRef
  35. 35.↵
    Wong J, Motulsky A, Abrahamowicz M, Eguale T, Buckeridge DL, Tamblyn R (2017) Off-label indications for antidepressants in primary care: descriptive study of prescriptions from an indication based electronic prescribing system. BMJ 356:j603. doi: 10.1136/bmj.j603
    OpenUrlAbstract/FREE Full Text
  36. 36.↵
    Reid RD, Pritchard G, Walker K, Aitken D, Mullen K-A, Pipe AL (2016) Managing smoking cessation. Can Med Assoc J 188:E484–E492. doi: 10.1503/cmaj.151510
    OpenUrlFREE Full Text
  37. 37.↵
    Chong MS, Bajwa ZH (2003) Diagnosis and treatment of neuropathic pain. J Pain Symptom Manage 25:S4–S11
    OpenUrlCrossRefPubMedWeb of Science
  38. 38.↵
    Gill JM, Klinkman MS, Chen YX (2010) Antidepressant medication use for primary care patients with and without medical comorbidities: a national electronic health record (EHR) network study. J Am Board Fam Med JABFM 23:499–508. doi: 10.3122/jabfm.2010.04.090299
    OpenUrlAbstract/FREE Full Text
  39. 39.↵
    Castro VM, Minnier J, Murphy SN, Kohane I, Churchill SE, Gainer V, Cai T, Hoffnagle AG, Dai Y, Block S, Weill SR, Nadal-Vicens M, Pollastri AR, Rosenquist JN, Goryachev S, Ongur D, Sklar P, Perlis RH, Smoller JW, Lee PH, Stahl EA, Purcell SM, Ruderfer DM, Charney AW, Roussos P, Pato C, Pato M, Medeiros H, Sobel J, Craddock N, Jones I, Forty L, DiFlorio A, Green E, Jones L, Dunjewski K, Landén M, Hultman C, Juréus A, Bergen S, Svantesson O, McCarroll S, Moran J, Chambert K, Belliveau RA (2015) Validation of electronic health record phenotyping of bipolar disorder cases and controls. Am J Psychiatry 172:363–372. doi: 10.1176/appi.ajp.2014.14030423
    OpenUrlCrossRefPubMed
  40. 40.↵
    Chen C-Y, Lee PH, Castro VM, Minnier J, Charney AW, Stahl EA, Ruderfer DM, Murphy SN, Gainer V, Cai T, Jones I, Pato CN, Pato MT, Landén M, Sklar P, Perlis RH, Smoller JW (2018) Genetic validation of bipolar disorder identified by automated phenotyping using electronic health records. Transl Psychiatry 8:86. doi: 10.1038/s41398-018-0133-7
    OpenUrlCrossRef
  41. 41.
    Kho AN, Pacheco JA, Peissig PL, Rasmussen L, Newton KM, Weston N, Crane PK, Pathak J, Chute CG, Bielinski SJ, Kullo IJ, Li R, Manolio TA, Chisholm RL, Denny JC (2011) Electronic Medical Records for Genetic Research: Results of the eMERGE Consortium. Sci Transl Med 3:79re1-79re1. doi: 10.1126/scitranslmed.3001807
    OpenUrlFREE Full Text
  42. 42.↵
    Zhou L, Baughman AW, Lei VJ, Lai KH, Navathe AS, Chang F, Sordo M, Topaz M, Zhong F, Murrali M, Navathe S, Rocha RA (2015) Identifying Patients with Depression Using Free-text Clinical Documents. Stud Health Technol Inform 216:629–633. doi: 10.3233/978-1-61499-564-7-629
    OpenUrlCrossRefPubMed
  43. 43.↵
    Carey DJ, Fetterolf SN, Davis FD, Faucett WA, Kirchner HL, Mirshahi U, Murray MF, Smelser DT, Gerhard GS, Ledbetter DH (2016) The Geisinger MyCode community health initiative: An electronic health record-linked biobank for precision medicine research. Genet Med 18:906–913. doi: 10.1038/gim.2015.187
    OpenUrlCrossRefPubMed
  44. 44.
    Borthwick K, Smelser D, Bock J, Elmore J, Ryer E, Ye Z, Pacheco J, Carrell D, Michalkiewicz M, Thompson W, Pathak J, Bielinski S, Denny J, Linneman J, Peissig P, Kho A, Gottesman O, Parmar H, Kullo I, McCarty C, Böttinger E, Larson E, Jarvik G, Harley J, Bajwa T, Franklin D, Carey D, Kuivaniemi H, Tromp G (2015) Ephenotyping for Abdominal Aortic Aneurysm in the Electronic Medical Records and Genomics (emerge) Network: Algorithm Development and Konstanz Information Miner Workflow. Int J Biomed Data Min 4:1–8. doi: 10.4172/2090-4924.1000113
    OpenUrlCrossRef
  45. 45.
    Maeng DD, Han JJ, Fitzpatrick MH, Boscarino JA (2017) Patterns of health care utilization and cost before and after opioid overdose: findings from 10-year longitudinal health plan claims data. Subst Abuse Rehabil 8:57–67. doi: 10.2147/SAR.S135884
    OpenUrlCrossRef
  46. 46.↵
    Boscarino JA, Kirchner HL, Pitcavage JM, Nadipelli VR, Ronquest NA, Fitzpatrick MH, Han JJ (2016) Factors associated with opioid overdose: a 10-year retrospective study of patients in a large integrated health care system. Subst Abuse Rehabil 7:131–141. doi: 10.2147/SAR.S108302
    OpenUrlCrossRef
  47. 47.↵
    Bhattarai N, Charlton J, Rudisill C, Gulliford MC (2013) Prevalence of depression and utilization of health care in single and multiple morbidity: a population-based cohort study. Psychol Med 43:1423–31. doi: 10.1017/S0033291712002498
    OpenUrlCrossRefPubMed
  48. 48.
    Felleman BI, Athenour DR, Ta MT, Stewart DG (2013) Behavioral health services influence medical treatment utilization among primary care patients with comorbid substance use and depression. J Clin Psychol Med Settings 20:415–26. doi: 10.1007/s10880-013-9367-y
    OpenUrlCrossRef
  49. 49.↵
    Possemato K, Bishop TM, Willis MA, Lantinga LJ (2013) Healthcare utilization and symptom variation among veterans using behavioral telehealth center services. J Behav Heal Serv Res 40:416–426. doi: 10.1007/s11414-013-9338-y
    OpenUrlCrossRef
  50. 50.↵
    Wen XJ, Wang LM, Liu ZL, Huang A, Liu YY, Hu JY (2014) Meta-analysis on the efficacy and tolerability of the augmentation of antidepressants with atypical antipsychotics in patients with major depressive disorder. Braz J Med Biol Res 47:605–16. doi: 10.1590/1414-431x20143672
    OpenUrlCrossRefPubMed
  51. 51.↵
    Verger P, Saliba B, Rouillon F, Kovess-Masféty V, Villani P, Bouvenot G, Lovell A (2008) Determinants of coprescription of anxiolytics with antidepressants in general practice. Can J Psychiatry 53:94–103. doi: 10.1177/070674370805300204
    OpenUrlCrossRefPubMed
  52. 52.↵
    Hawton K, Casañas i Comabella C, Haw C, Saunders K (2013) Risk factors for suicide in individuals with depression: A systematic review. J Affect Disord 147:17–28. doi: 10.1016/j.jad.2013.01.004
    OpenUrlCrossRefPubMed
  53. 53.↵
    Weitz E, Hollon SD, Kerkhof A, Cuijpers P (2014) Do depression treatments reduce suicidal ideation? The effects of CBT, IPT, pharmacotherapy, and placebo on suicidality. J Affect Disord 167:98–103. doi: 10.1016/j.jad.2014.05.036
    OpenUrlCrossRef
  54. 54.↵
    Wulsin LR, Vaillant GE, Wells VE (1999) A systematic review of the mortality of depression. Psychosom Med 61:6–17
    OpenUrlAbstract/FREE Full Text
  55. 55.↵
    Schulz R, Drayer RA, Rollman BL (2002) Depression as a risk factor for non-suicide mortality in the elderly. Biol Psychiatry 52:205–25. doi: 10.1016/S0006-3223(02)01423-3
    OpenUrlCrossRefPubMedWeb of Science
  56. 56.↵
    Moussavi S, Chatterji S, Verdes E, Tandon A, Patel V, Ustun B (2007) Depression, chronic diseases, and decrements in health: results from the World Health Surveys. Lancet 370:851–858. doi: 10.1016/S0140-6736(07)61415-9
    OpenUrlCrossRefPubMedWeb of Science
  57. 57.↵
    Deyo RA, Cherkin DC, Ciol MA (1992) Adapting a clinical comorbidity index for use with ICD-9-CM administrative databases. J Clin Epidemiol 45:613–619. doi: 10.1016/0895-4356(92)90133-8
    OpenUrlCrossRefPubMedWeb of Science
  58. 58.↵
    Kroenke K, Spitzer RL, Williams JBW (2001) The PHQ-9: Validity of a brief depression severity measure. J Gen Intern Med 16:606–613. doi: 10.1046/j.1525-1497.2001.016009606.x
    OpenUrlCrossRefPubMedWeb of Science
  59. 59.↵
    Manea L, Gilbody S, McMillan D (2015) A diagnostic meta-analysis of the Patient Health Questionnaire-9 (PHQ-9) algorithm scoring method as a screen for depression. Gen Hosp Psychiatry 37:67–75. doi: 10.1016/j.genhosppsych.2014.09.009
    OpenUrlCrossRefPubMed
  60. 60.↵
    Wray NR, Ripke S, Mattheisen M, Trzaskowski M, Byrne EM, Abdellaoui A, Adams MJ, Agerbo E, Air TM, Andlauer TMF, Bacanu S-A, Bækvad-Hansen M, Beekman AFT, Bigdeli TB, Binder EB, Blackwood DRH, Bryois J, Buttenschøn HN, Bybjerg-Grauholm J, Cai N, Castelao E, Christensen JH, Clarke T-K, Coleman JIR, Colodro-Conde L, Couvy-Duchesne B, Craddock N, Crawford GE, Crowley CA, Dashti HS, Davies G, Deary IJ, Degenhardt F, Derks EM, Direk N, Dolan C V., Dunn EC, Eley TC, Eriksson N, Escott-Price V, Kiadeh FHF, Finucane HK, Forstner AJ, Frank J, Gaspar HA, Gill M, Giusti-Rodríguez P, Goes FS, Gordon SD, Grove J, Hall LS, Hannon E, Hansen CS, Hansen TF, Herms S, Hickie IB, Hoffmann P, Homuth G, Horn C, Hottenga J-J, Hougaard DM, Hu M, Hyde CL, Ising M, Jansen R, Jin F, Jorgenson E, Knowles JA, Kohane IS, Kraft J, Kretzschmar WW, Krogh J, Kutalik Z, Lane JM, Li Y, Li Y, Lind PA, Liu X, Lu L, MacIntyre DJ, MacKinnon DF, Maier RM, Maier W, Marchini J, Mbarek H, McGrath P, McGuffin P, Medland SE, Mehta D, Middeldorp CM, Mihailov E, Milaneschi Y, Milani L, Mill J, Mondimore FM, Montgomery GW, Mostafavi S, Mullins N, Nauck M, Ng B, Nivard MG, Nyholt DR, O’Reilly PF, Oskarsson H, Owen MJ, Painter JN, Pedersen CB, Pedersen MG, Peterson RE, Pettersson E, Peyrot WJ, Pistis G, Posthuma D, Purcell SM, Quiroz JA, Qvist P, Rice JP, Riley BP, Rivera M, Saeed Mirza S, Saxena R, Schoevers R, Schulte EC, Shen L, Shi J, Shyn SI, Sigurdsson E, Sinnamon GBC, Smit JH, Smith DJ, Stefansson H, Steinberg S, Stockmeier CA, Streit F, Strohmaier J, Tansey KE, Teismann H, Teumer A, Thompson W, Thomson PA, Thorgeirsson TE, Tian C, Traylor M, Treutlein J, Trubetskoy V, Uitterlinden AG, Umbricht D, Van der Auwera S, van Hemert AM, Viktorin A, Visscher PM, Wang Y, Webb BT, Weinsheimer SM, Wellmann J, Willemsen G, Witt SH, Wu Y, Xi HS, Yang J, Zhang F, Arolt V, Baune BT, Berger K, Boomsma DI, Cichon S, Dannlowski U, de Geus ECJ, DePaulo JR, Domenici E, Domschke K, Esko T, Grabe HJ, Hamilton SP, Hayward C, Heath AC, Hinds DA, Kendler KS, Kloiber S, Lewis G, Li QS, Lucae S, Madden PFA, Magnusson PK, Martin NG, McIntosh AM, Metspalu A, Mors O, Mortensen PB, Müller-Myhsok B, Nordentoft M, Nöthen MM, O’Donovan MC, Paciga SA, Pedersen NL, Penninx BWJH, Perlis RH, Porteous DJ, Potash JB, Preisig M, Rietschel M, Schaefer C, Schulze TG, Smoller JW, Stefansson K, Tiemeier H, Uher R, Völzke H, Weissman MM, Werge T, Winslow AR, Lewis CM, Levinson DF, Breen G, Børglum AD, Sullivan PF (2018) Genome-wide association analyses identify 44 risk variants and refine the genetic architecture of major depression. Nat Genet 50:668–681. doi: 10.1038/s41588-018-0090-3
    OpenUrlCrossRefPubMed
  61. 61.↵
    Euesden J, Lewis CM, O’Reilly PF (2015) PRSice: Polygenic Risk Score software. Bioinformatics 31:1466–1468. doi: 10.1093/bioinformatics/btu848
    OpenUrlCrossRefPubMed
  62. 62.↵
    Hamet P, Tremblay J (2005) Genetics and genomics of depression. Metabolism 54:10–15. doi: 10.1016/j.metabol.2005.01.006
    OpenUrlCrossRefPubMedWeb of Science
  63. 63.↵
    Kamiya Y, Doyle M, Henretta JC, Timonen V (2013) Depressive symptoms among older adults: the impact of early and later life circumstances and marital status. Aging Ment Health 17:349–57. doi: 10.1080/13607863.2012.747078
    OpenUrlCrossRefPubMedWeb of Science
  64. 64.↵
    Yan X-Y, Huang S-M, Huang C-Q, Wu W-H, Qin Y (2011) Marital Status and Risk for Late Life Depression: A Meta-Analysis of the Published Literature. J Int Med Res 39:1142–1154. doi: 10.1177/147323001103900402
    OpenUrlCrossRefPubMed
  65. 65.↵
    Kessing LV, Willer I, Andersen PK, Bukh JD (2017) Rate and predictors of conversion from unipolar to bipolar disorder: A systematic review and meta-analysis. Bipolar Disord 19:324–335. doi: 10.1111/bdi.12513
    OpenUrlCrossRef
  66. 66.↵
    Gallo JJ, Hwang S, Joo JH, Bogner HR, Morales KH, Bruce ML, Reynolds CF (2016) Multimorbidity, Depression, and Mortality in Primary Care: Randomized Clinical Trial of an Evidence-Based Depression Care Management Program on Mortality Risk. J Gen Intern Med 31:380–386. doi: 10.1007/s11606-015-3524-y
    OpenUrlCrossRef
  67. 67.↵
    Sinnige J, Braspenning J, Schellevis F, Stirbu-Wagner I, Westert G, Korevaar J (2013) The prevalence of disease clusters in older adults with multiple chronic diseases - A systematic literature review. PLoS One 8:. doi: 10.1371/journal.pone.0079641
    OpenUrlCrossRefPubMed
  68. 68.↵
    CMS (2015) CMS Chronic Conditions Data Warehouse (CCW) CCW Condition Algorithms. Chronic Cond Algorithms 72:2015. doi: 10.1002/pds.2329/pdf
    OpenUrlCrossRef
  69. 69.↵
    ICD10Data.com Convert ICD-9-CM Codes to ICD-10-CM codes. https://www.icd10data.com/Convert
Back to top
Posted October 17, 2018.
Download PDF
Email
Defining Major Depressive Disorder Cohorts Using the EHR: Multiple Phenotypes Based on ICD-9 Codes and Medication Orders
Wendy Marie Ingram, Anna M. Baker, Christopher R. Bauer, Jason P. Brown, Fernando S. Goes, Sharon Larson, Peter P. Zandi
bioRxiv 227561; doi: https://doi.org/10.1101/227561
Twitter logo Facebook logo LinkedIn logo Mendeley logo
Citation Tools

Subject Area