Ameena T. Ahmed, MD, MPH

 To examine the patient and provider characteristics associated with utilization of mental health services (MHS) among women experiencing intimate partner violence (IPV).

 Cross-sectional study among 6870 women aged 18 to 65 years with first IPV identification between May 2004 and December 2009 in Kaiser Permanente Northern California.

 Utilization of MHS within 60 days after first IPV identification was determined. Multivariate generalized estimating equation logistic regression models that controlled for patient and provider characteristics were used to determine predictors of utilization.

 Thirty-seven percent of women utilized MHS. In multivariate generalized estimating equation models, the strongest predictor of utilization was electronic referral (odds ratio [OR], 4.40; 95% confidence interval [CI], 3.66-5.28). Odds of utilization were lower among black (OR, 0.71; 95% CI, 0.57-0.89), Latina (OR, 0.62; 95% CI, 0.41-0.95), and Spanish-speaking (OR, 0.71; 95% CI, 0.57-0.89) patients and were higher among those with prior posttraumatic stress disorder (OR, 2.38; 95% CI, 1.17-3.44) or depression (OR, 1.35; 95% CI, 1.17-1.57). Emergency department identification of IPV was associated with lower odds of MHS utilization (OR, 0.46; 95% CI, 0.37-0.59), while older provider identification of IPV was associated with higher odds of MHS utilization (OR, 1.33; 95% CI, 1.07-1.65).

 Additional training for providers, particularly those who are younger or are practicing in emergency departments, may be needed to increase rates of MHS utilization among patients affected by IPV. Addressing language barriers to care and cultural appropriateness may improve MHS utilization.

Intimate partner violence is a common event in women's lives.

Lower rates of utilization of mental health services among women after intimate partner violence diagnosis by younger providers and in emergency department settings suggest these as points of intervention to improve referral rates. 

Utilization of mental health services is lower among nonwhite and non-English-speaking patients, suggesting language barriers to care and possibly the need to tailor intimate partner violence programs to minority patients.

	Intimate partner violence (IPV) is a common experience in the lives of women

 and is associated with deleterious effects on physical and mental health

 Professional medical organizations support routine IPV screening in medical care settings

 and recommend that clinicians offer support and provide information about safety planning and community-based advocacy resources to IPV-affected patients.

 The implementation of a coordinated approach to IPV prevention has demonstrated a dramatic and sustained increase in identification, intervention, and referral in a large healthcare delivery system.

 Provision of follow-up referral for mental health or social services is often suggested, as women who are experiencing IPV also experience psychological distress

Assessment of  utilization patterns is important because the ultimate goal of IPV identification is the receipt of assistance that the patient can use to address her mental health needs, improve her safety, and change her situation. However, few data are available about the frequency or predictors of utilization of MHS among IPV-affected women. The objective of this study was to determine the predictors of women’s utilization of MHS in the managed care setting.

	This cross-sectional study was conducted among the membership of Kaiser Permanente Northern California Medical Services Plan (hereafter, Kaiser), an integrated healthcare delivery system that provides comprehensive care, including MHS, for approximately 3.1 million members in a 14-county region of northern California. Kaiser membership comprises almost one-third of the population in the catchment area and is representative of the underlying population, except for slightly lower membership at the extremes of income. Its racial/ethnic composition is similar to that of the US census—enumerated population in the San Francisco Bay metropolitan statistical area, namely, 7% African American in the Kaiser membership and 9% African American in the underlying metropolitan statistical area population, with 10% and 11% Asian or Pacific Islander, 10% and 14% Hispanic, and 71% and 64% white, respectively. Since 2002, Kaiser providers have had hard-copy and Internet access to clinical practice recommendations on screening, identification, and referral of IPV-affected patients. Training and continuing medical education credit has been offered annually at most Kaiser facilities. In addition, each clinical facility has an IPV physician champion, and most have a multidisciplinary team that provides facility clinician education and oversees IPV quality improvement efforts. Patient education materials are available in all clinics. This study was approved by the Kaiser Foundation Research Institute Institutional Review Board.

	After excluding 4277 women whose first IPV identification (ie, 

International Classification of Diseases, Ninth Revision, Clinical Modification

 visit code) occurred in mental health clinics, our study sample comprised 6870 adult women aged 18 to 65 years with first IPV identification between May 2004 and December 2009. May 2004 was the month during which all Kaiser facilities implemented electronic referrals. To ensure that we recorded the first identification of IPV within Kaiser, we excluded women with prior IPV identification any time after January 2000 or who did not have at least 1 year of continuous Kaiser membership before identification. Furthermore, because the probability of utilization among inpatient referrals is 100% (because the physician comes to see the patient while hospitalized), we excluded inpatient IPV codes.

	Kaiser maintains an electronic medical record in which each patient is identified by a unique medical record number. The electronic medical record includes patient demographic information, clinic visit diagnoses, and the provider associated with each visit. Since May 2004, an electronic referral system has allowed the referring provider to enter the service to which a patient is referred and the reason for referral; referring provider information is automatically populated into the referral, which is then delivered electronically.

	We defined the index date as the date of first outpatient or emergency department (ED) IPV identification. The outcome, mental health utilization (hereafter, utilization), was defined as a patient attending an MHS visit (including psychiatry, behavioral medicine, or substance use treatment) within 60 days after the index date.

	Putative confounding or explanatory variables were assessed at patient and provider levels. Patient age, sex, self-reported race/ethnicity, and primary language were obtained from administrative databases. Patient psychiatric comorbidities (posttraumatic stress disorder [PTSD] or depression) diagnosed during the 2 years before the index date at any outpatient visit were assessed from electronic records. Prior utilization was defined as any utilization during the 12 months before the index date. Provider age, race/ ethnicity, and sex were determined from medical group records. The diagnosing clinic was defined as primary care (internal medicine or family practice), obstetrics and gynecology (OB/GYN), ED (or urgent care), or other. Language concordance was defined as the provider speaking the patient’s preferred language. Mental health services referral (hereafter, e-referral) was defined as entry of electronic referral by the diagnosing provider within 60 days after the index date. Within Kaiser, members may access MHS directly by calling the mental health intake line, as well as through provider-initiated referral. As shown in the 

, our definition is conservative, as it excludes cases in which providers called the MHS intake line with the patient in the office, advised patients to call MHS on their own, or escorted the patient to colocated MHS providers in primary care clinics.

	We calculated the frequency of utilization among the entire cohort and in strata defined by referral and by patient and provider characteristics. The c2 test was used to examine differences in utilization by level of each characteristic.

	We then used multivariate generalized estimating equation (GEE) logistic regression models that included patient and provider variables to determine which patient or provider factors explained the differences observed in utilization. We used the GEE method to account for clustering (correlation) of data by provider. Some providers identified multiple IPV-affected women. Therefore, provider characteristics (age, sex, race/ ethnicity, and clinic type) may be correlated within providers (concordance). Our outcome (ie, how likely a provider is to ensure that IPV-affected women are referred to and use MHS) may also be correlated within providers. The GEE provides a way to build a generalized linear model using correlated data and returning robust standard errors, even when the covariance matrix specified in the model is not correctly specified.19

	Patient-level variables included age, race/ethnicity (white, black, Latina, Asian, or other or missing), primary language (English, Spanish, or other), prior mental health diagnosis (PTSD with or without depression, depression only, or none), and prior utilization. Provider-level variables included provider age (quartiles), sex, race/ethnicity (white, black, Latina,  Asian, or other or missing), clinic type (primary care, OB/GYN, ED, or other), and e-referral. In addition, models included dummy variables for calendar year of IPV identification to control for any temporal trends.

	Because odds ratios (ORs) do not approximate the risk ratio when the outcome is common, as is the case in our study, we also calculated predicted probabilities for each category of each covariate, using least squares means. We present these results to facilitate interpretation of the predicted difference in probability of the outcome (utilization) in multivariate models. Commercially available statistical software (SAS, version 9.1; SAS Institute Inc, Cary, NC) was used for all analyses.

Among 6870 adult women with first IPV identification between May 2004 and December 2009, the mean (SD) age was 37.8 (11.8) years, and there was significant racial/ethnic  diversity (

). Almost 90% spoke English as their primary language, whereas 8% spoke Spanish and 2% other languages. A language-concordant physician identified 92%. Prior mental health diagnoses were common, with 25% having prior PTSD or depression. The mean age of diagnosing providers was 46 years; 64% of diagnosing providers were female. Most IPV-affected women were identified by providers working in primary care (48%), with fewer in OB/GYN (30%) or EDs (18%). The identifying provider performed e-referral to MHS within 60 days of IPV identification in 849 (12%) of 6870 women.

Among 6870 women with first IPV identification, 2527 (37%) used MHS, including 550 patients with e-referral and 1977 patients without e-referral (

). All patient and provider characteristics were significantly associated (

 <.003) with utilization among the entire cohort and among patients without e-referral. For example, among the entire cohort, having a prior PTSD (61%) or depression (46%) diagnosis was associated with higher probability of utilization than having no previously diagnosed mental health comorbidity (33%). English-speaking patients (37%) were more likely to use MHS than those who spoke Spanish (32%) or other languages (28%), and patient—physician language concordance was associated with a 6% absolute higher utilization frequency. Patient age was associated with utilization, with highest utilization among patients aged 45 to 65 years (40%) and lowest utilization among those aged 25 to 34 years (34%). Among provider characteristics, older provider age was associated with higher frequency of utilization (45% vs 37% for the youngest quartile, 

 <.001). Patients of female providers (40%) were more likely to use MHS than those of male providers (32%). Utilization was higher among patients diagnosed in primary care (41%) and OB/GYN (37%) clinics than among those diagnosed in EDs (21%).

	Among patients with e-referral, utilization varied by patient race/ethnicity (

 = .06), with white women most likely (72%) and Latinas least likely (58%) to use MHS (Table 2). Provider age was associated with utilization (

 = .05), being highest among providers aged 46 to 54 years (69%) and lowest among providers younger than 37 years (62%). Utilization was highest among cases identified in primary care (68%), followed by OB/GYN (63%) and EDs (53%) (

 = .007). Although the associations of prior mental health diagnosis, patient language, patient—provider language concordance, and provider sex did not reach statistical significance, the absolute differences in rates of utilization between levels of each of these variables were similar to differences among patients without e-referral.

	In multivariate GEE models of utilization that controlled for clustering of observations by provider (

), e-referral was the strongest predictor of utilization (OR, 4.40; 95% confidence interval [CI], 3.66-5.28). Odds of utilization were lower among blacks (OR, 0.71; 95% CI, 0.57-0.89) and Latinas (OR, 0.62; 95% CI, 0.41-0.95) compared with whites. Furthermore, Spanish speakers had lower odds of utilization compared with English speakers (OR, 0.71; 95% CI, 0.57-0.89). Odds of utilization were higher among patients with prior mental health morbidity. Patients identified by the oldest 2 quartiles of providers had modestly increased odds of utilization (OR, 1.33; 95% CI, 1.07-1.65 for the oldest providers). Utilization was not significantly different between those identified in Ob/gyn settings compared with those identified in primary care, but identification in EDs was associated with an OR for utilization of 0.46 (95% CI, 0.37-0.59).

	As expected, we found similar point estimates but wider CIs using GEE models vs using multivariate logistic regressionmodels that did not account for clustering of observations.

	The finding that each association remained significant when using GEE models suggests that the findings are robust.

	Among 6870 adult women with first identification of IPV, 37% used MHS within 60 days of identification. In multivariate models that included patient-level and provider-level variables, receiving provider-initiated electronic referral was the strongest predictor of utilization. Odds of utilization were lower among black and Latina patients, Spanish-speaking patients, and those identified in EDs. Provider demographic characteristics were unassociated with utilization, other than a modest association with older provider age.

	To the best of our knowledge, this is the first published study to assess predictors of MHS utilization among IPV-affected women identified in the healthcare setting. It is also the only study to date offering information on both provider and patient characteristics associated with utilization. The literature on referral rates of patients with other types of trauma is sparse. In one study

 of patients hospitalized because of traumatic physical injury, 19% received MHS referral, and 51% of those used MHS. These data were based on patient self-report and excluded patients with injuries caused by domestic violence. The lower referral rate in our study (12%) likely reflects our conservative definition of referral (Figure). The high utilization rate among referred patients may reflect the straightforward patient access to MHS within Kaiser.

 examined patient factors related to help-seeking behavior, including obtaining MHS, among IPV-affected patients. In a study

	factors such as patient need (symptoms causing distress) and predisposing characteristics (age, sex, race/ethnicity, and language) were found to be important in determining receipt of MHS treatment. Patients experiencing high levels of distress were more likely to receive treatment. Our study may reflect similar patterns of help-seeking behavior, as patients with prior diagnosis of PTSD or depression were more likely to use MHS.

	There are several potential explanations for our finding that women who were Spanish speaking, black, or Latina were less likely than English-speaking or white women to use MHS. It was previously found that the most common reasons for IPV-affected women to accept referral were unhappiness with their situation, fear for their physical well-being, concern about their children who witnessed abuse, and the suggestion by a healthcare provider that their symptoms could be related to the abuse.

 Although many persons who experience mental health disorders do not seek treatment, this is particularly true among minority populations.

 Racial/ethnic minorities may differ in their tendency to associate mental health problems with their symptoms. If clinicians do not discuss connections between IPV and a patient’s symptoms, patients may not anticipate benefit in MHS follow-up. One study

 of patients with depression showed less clinician—patient communication and rapport building among some racial/ethnic minority patients. This could also affect the likelihood of patients’ using MHS. In addition, minority or non–English-speaking groups may be more likely than white English-speaking patients to choose other support networks (eg, clergy or advocacy agencies) as alternatives to MHS.

 It is possible that women with prior mental health diagnosis were more likely to accept care because of higher levels of psychological distress. Alternately, higher utilization might reflect return to a previously established mental health provider.

	Our study has several limitations. First, some women who experience IPV are never identified in the medical setting. Our study did not investigate this phenomenon. Second, our definition of referral is conservative, as we were unable to determine which women were referred through in-clinic telephone calls for immediate MHS intake or through a provider’s verbal directions to call the MHS intake line. Third, we lack data to identify women who obtained MHS or social services outside of Kaiser. Our estimation of predictors of utilization would be biased if certain groups of women were more likely to seek outside services. However, because MHS within Kaiser are a covered benefit for most members, there is a cost incentive for women to use services within Kaiser. Fourth, we did not collect information from women about why they did or did not use MHS. Fifth, data on sex, race/ethnicity, and age are missing from 15% of diagnosing providers. Almost all of these were working per diem in EDs. Sixth, we lacked data on duration, type (physical, sexual, or emotional), and severity of IPV. It is possible that these characteristics affected the likelihood of utilization.

	Our study has multiple strengths. First, we present data from a large cohort of women identified as IPV survivors at Kaiser, which provides medical care to more than 3 million individuals. Our findings have applicability to a large segment of the population. Second, our electronic databases allow complete ascertainment of clinically recognized IPV, electronic referral, and utilization within Kaiser. Our ability to fully capture utilization data obviates the possibility of recall bias.

	Our findings suggest several potential targets for interventions to increase rates of utilization among women experiencing IPV who present to the healthcare system. First, there is a particular need to focus on improving rates of provider referral, as the present study and previous investigations have found that provider referral has the strongest relationship to utilization, far outweighing that of patient characteristics.

 Ongoing training for providers about what the health consequences of IPV are and how to refer to appropriate services may improve referral rates. Second, interventions need to focus on increasing utilization among non—English-speaking and nonwhite women. Although providers are more likely to refer these patients, they are less likely to use services. This suggests that addressing language barriers to care may be a useful intervention. Alternately, these patients may prefer community-based rather than medical system–based care. Third, lower utilization among black and Latina women suggests that services might be, or might be perceived to be, culturally inappropriate. Fourth, additional training of providers, particularly those who are younger or are practicing in EDs, might increase rates at which providers refer IPV survivors to MHS. After controlling for e-referral, utilization did not differ between OB/GYN and primary care clinics, although it remained significantly lower for EDs.

	Additional study should determine the comparative effectiveness of strategies to improve provider referral, which is primarily dependent on provider knowledge and behavior, as well as how to ensure timely access to culturally and linguistically appropriate services. Furthermore, future community-based participatory research might best determine how to deliver appropriate services to members of racial/ethnic, cultural, or linguistic minorities and to women of different ages. In particular, it will be important to further examine MHS utilization by race/ethnicity and by other social factors. There is a need to interview women who do not use MHS after referral, to determine whether they are seeking medical services outside of the Kaiser system, obtaining services from community-based organizations, or seeking help from social networks, including family, friends, or clergy. It is important to determine if these findings can be duplicated in other healthcare systems that have implemented a coordinated approach to IPV prevention.

Articles

Among women diagnosed as having intimate partner violence, odds of mental health utilization were lower among those diagnosed in emergency departments versus primary care clinics.

Mental Health Services Utilization Among Women Experiencing Intimate Partner Violence

 To study the effect of cost-sharing policy changes on utilization of test strips for self-monitoring of blood glucose.

 A legislative mandate (January 1, 2000) required California health plans to cover diabetes supplies, including those for self-monitoring of blood glucose. One health plan, Kaiser Permanente Northern California, initially waived established copayments and provided free test strips to members with diabetes mellitus for 2 years but later instituted a 20% coinsurance charge for a portion of their membership. 

 A retrospective cohort design was used to study pharmacy-based estimates of test strip utilization changes during this natural experiment. Analyses included 2 cohort investigations using pretest-posttest analysis with control subjects to study transitions from a copayment period to a free test strip period and from the free test strip period to a coinsurance period.

 During the copayment period, test strip utilization was inversely related to copayments for test strips. Offering free test strips did not increase utilization, even among those paying higher copayments before the policy change. Price-elastic patterns formed before and during the copayment period persisted, despite receiving free test strips for 2 years. The coinsurance, introduced after 2 years of receiving free test strips, resulted in statistically significant (but not clinically relevant) decreased utilization (approximately 1-3 fewer test strips/month). Change patterns did not differ by socioeconomic status.

 Offering free test strips shifted costs from patient to health plan, without improving adherence. The introduced coinsurance slightly reduced utilization and adherence to recommendations about self-monitoring of blood glucose. Neither intervention had marked clinical effect. Cross-sectional analyses should not be used to predict utilization changes in the face of rapidly evolving benefit policies.

We evaluated changes in utilization of test strips for self-monitoring of blood glucose levels across the following 2 cost-sharing thresholds: (1) from a period of copayments to a period of free strips and then (2) from the period of free strips to a period of coinsurance.

Findings included slightly reduced utilization in response to increased cost sharing but unchanged utilization in response to free test strips.

Price-elastic patterns formed before and during the period of copay-ments persisted, despite receiving free test strips for 2 years.

Cross-sectional analyses should not be used to predict utilization changes in response to changes in benefit policies.

The Diabetes Control and Complications Trial

and the United Kingdom Prospective Diabetes Study

demonstrated the benefits of achieving tight glycemic control in patients with diabetes mellitus (DM) and led to a concomitant promotion of self-monitoring of blood glucose (SMBG) as an integral part of good diabetes care. However, SMBG is consistently practiced less than is recommended by commonly supported guidelines,

including those of the American Diabetes Association.

Test strips for SMBG are expensive, and SMBG practice is more sensitive to cost sharing

(ie, price-elastic demand) than many other health services (eg, dilated eye examinations).

However, current understanding regarding how financial barriers affect utilization is primarily based on cross-sectional analyses.

California Senate Bill 64 (SB64), implemented on January 1, 2000, mandated that California's health plans provide coverage for diabetes supplies, including those for SMBG. The bill was intended to eliminate a financial barrier to SMBG and to facilitate this self-care practice. In recent years, 37 other states have passed similar legislation. Following similar legislation, rates of SMBG utilization increased, and glycemic control improved after a New England health plan started providing free glucose monitors in 1993.

In the year before implementation of SB64 (hereafter referred to as the copayment period), a California health plan, Kaiser Permanente Northern California (KPNC), charged members with DM a $0 to $30 copayment for each prescription of test strips, depending on the member's type of benefit coverage. During the 2 years after SB64 was implemented (January 1, 2000, through December 31, 2001) (hereafter referred to as the free test strip period), test strips were provided free of charge to all KPNC members with DM. A second policy change occurred on January 1, 2002, after which approximately 35% of the commercial membership were charged a 20% coinsurance (approximately $8-$12 per prescription, depending on the brand of test strips) (hereafter referred to as the coinsurance period).

We studied the effects of these 2 benefit policy changes on SMBG practice (as measured by test strip utilization). We hypothesized an increase in individual-level test strip utilization after test strips were provided for free, with the greatest increase among those paying the largest copayments before the implementation of SB64. We assumed that poorer members would be more price sensitive and hypothesized that provision of free test strips would reduce socioeconomic disparities in utilization. We also hypothesized that, after receiving free test strips for 2 years, utilization would decrease after the coinsurance was introduced and that this decrease would be greatest among the poor and among those not treated with insulin.

Kaiser Permanente Northern California is a group-model healthcare organization that provides comprehensive medical services to more than 3 million members (approximately 35% of the population in the surrounding geographic area) through 15 hospitals and 23 outpatient clinics. Study subjects were identified from the KPNC diabetes registry (approximately 132 000 in 2000).

Two cohorts were created to analyze the effect of the 2 separate policy changes. The first was created for the analysis of utilization before and after the introduction of free test strips (hereafter referred to as the free test strip cohort), and the second was created to assess utilization before and after institution of a coinsurance for test strips (hereafter referred to as the coinsurance cohort). For each cohort, we created a sampling frame that included subjects with continuous health plan membership and prescription drug benefits in the prepolicy and postpolicy periods who were also ongoing test strip users (defined as having at least 1 dispensing for test strips during each year of the study). We did not include members without prescription drug benefits and those with prescription drug caps, because they have little financial incentive to purchase SMBG supplies exclusively at KPNC pharmacies and we cannot ascertain utilization at non-KPNC pharmacies. Based on these inclusion criteria, we identified 33 596 subjects in the free test strip cohort and 27 810 subjects in the coinsurance cohort sampling frames. We then excluded 12 241 and 9841 noncommercial health plan members (ie, members in subsidized plans [eg, Medicare]) from the free test strip cohort and the coinsurance cohort, respectively, to reduce case mix, demographic, and socioeconomic heterogeneity in the cohort. Patients enrolled in Medicare received free test strips starting July 1, 1998; this age class would have dominated the reference group if included.

Pharmacy medication dispensing records were then used to classify subjects into the following 3 mutually exclusive treatment groups: patients who use any regimen that includes insulin (hereafter referred to as the insulin group), patients treated only with oral hypoglycemic agents (OHAs) (hereafter referred to as the OHA group), and patients receiving no diabetes pharmacotherapy but receiving medical nutrition therapy (MNT) only (hereafter referred to as the MNT group). Healthcare providers generally encourage increases in SMBG frequency with the intensification of or the introduction of new pharmacotherapy; in these individuals, we would expect an increase in SMBG frequency during this time, independent of the change in the cost of test strips. Therefore, we excluded 2107 and 2323 additional subjects from the free test strip cohort and the coinsurance cohort, respectively, who switched type of diabetes therapy modalities (eg, adding insulin to regimens that previously included oral agents only) during the study period, although we could not exclude those with dosing changes or with additions of other therapies within the oral agent or insulin modality. The final free test strip cohort and coinsurance cohort on which this study was based included 19 248 and 15 646 subjects, respectively.

Because the 2 benefit changes in charges for test strips were not announced in writing to the KPNC membership, only ongoing users were aware of price changes (via the change in out-of-pocket charge for test strips). Therefore, for our main analyses of the individual-level effect of changes in out-of-pocket cost (eg, pre-post policy), we included only test strip users at baseline whom we assumed had experienced the changes in charges for test strips. However, although exclusion of individuals who were not prevalent users was appropriate for assessing individual-level changes, these cohorts are not well suited to study population-level patterns of utilization across levels of out-of-pocket cost because high cost may also lead to never initiating use or to discontinuing use. Therefore, we evaluated trends in SMBG practice and price sensitivity in a panel study of annual cohorts that included the individuals who were nonusers.

We assessed SMBG practice as indicated by test strip utilization (mean daily test strip utilization) recorded from pharmacy dispensing records. The use of pharmacy dispensing records to estimate utilization has been shown to be a valid measure for patients with a prescription benefit.

For each calendar year, the mean number of test strips used per day was calculated for each member by summing the cumulative number of test strips for the year and dividing by 365 days. We assumed that a patient possessed no stockpile of test strips at baseline. Test strip copayment amounts paid out-of-pocket were available through KPNC's benefits database. Copayment levels were collapsed into the following 4 categories: $0 (reference group), $1 to $3, $4 to $9, and $10 and higher to identify price-elastic patterns. Neighborhood-level socioeconomic status was characterized by geocoding each member's address to its 2000 census block group and by creating variables for median household income in 1999 and for residence in an impoverished neighborhood (defined as >20% of the population with an income below the US poverty line).

We first detailed population-level 1995 to 2002 secular trends in SMBG (unadjusted), defined as a continuous measure (mean number of test strips used per day) and as a discrete measure (prevalent use vs nonuse). We then used a longitudinal retrospective cohort investigation in 2 separate cohorts to evaluate longitudinal changes in test strip utilization spanning a 4-year period across the following 2 thresholds: (1) transition from the copayment period to the free test strip period and then (2) transition from the free test strip to the coinsur-ance period. The outcome of interest was the change in the mean number of test strips used per day associated with the policy-based health plan—wide change in cost sharing. To control for selection bias and for intervening time factors (secular trends and aging), we used the difference-in-differences method (pretest-posttest analysis with control subjects).

Hierarchical linear models (random coefficient models) were specified to determine the 1-year change in the mean daily test strip utilization following the introduction of 2 separate cost-sharing policies. All statistical analyses were conducted using SAS software (SAS Institute, Cary, NC); the multivariate models were run using PROC MIXED (SAS Institute). The reference group, the $0 copayment group, received free test strips before and after each policy change and was used to account for the secular trend in SMBG in the absence of any price change. We expected that the levels of SMBG and the patterns of change in test strip utilization might cluster within the different medical facilities, which may have differed in their efforts to promote SMBG; therefore, we specified a random intercept for medical facility in our statistical models.

We assumed a priori whose effects would be modified by diabetes therapy, so we specified separate models for the insulin, OHA, and MNT groups. The exposures of interest (benefit-based changes in test strip copayment) were exogenous policy changes and were unlikely to be related to any individual-level factors. To be sure, we also specified adjusted models that controlled for potentially confounding individual-level prebaseline variables. These covariates included the mean daily test strip utilization during the year before the policy change, prepolicy prescription drug copayment (for the analysis of the first policy change only), prepolicy glycosylat-ed hemoglobin level, age, missed appointment rate, sex, inpa-tient comorbidity score, total pharmacy cost in the prebaseline year, specialty visits, urgent care visits, primary care visits, residence in a predominantly working-class occupation neighborhood, residence in a predominantly undered-ucated neighborhood, block group median income (1999), oral medication refill adherence (OHA models only), and pharmacy-derived utilization of insulin syringes as a measure of daily insulin injection frequency (insulin models only). We found no substantive difference between unadjusted and adjusted models; therefore, we present only the unadjusted findings. In addition, we specified models with cross-product terms stratified by contextual measures of socioeconomic status, as we hypothesized that residence in an impoverished area might modify the degree of changes in SMBG after the implementation of each policy.

Price Elasticity and Population Secular Trends

We evaluated the population-level secular trends in SMBG as a discrete measure (prevalent use vs nonuse) (

) and as a continuous measure (mean number of test strips used per day) (data not shown) between 1995 and 2002 using the cohorts that did not exclude nonusers or inconsistent test strip users. There was a gradual trend toward higher prevalence of SMBG and more test strips used per day during the observation window. Consistent with price-elastic utilization, the mean daily test strip use and the prevalence of SMBG were consistently lower among those paying the higher out-of-pocket copayment at any point in time. Moreover, although the copayment charges that applied during the copayment period (1995-1999) were no longer in effect after 1999, the price-elastic patterns formed during the copayment period persisted.

Baseline Characteristics of the Free Test Strip Cohort

We identified 19 248 members with DM in the free test strip cohort who had pharmacy benefits, were test strip users during the year preceding the implementation of SB64, and did not change diabetes therapy type during the follow-up period. The reference group for the free test strip cohort comprised approximately 7% of the subjects who received test strips for free before the policy change (

). The remaining subjects were distributed across different levels of copay-ment. The free test strip cohort included the following 3 treatment strata: 8151 subjects in the insulin group, 10 049 subjects in the OHA group, and 1048 subjects in the MNT group. Given the large sample sizes, it is important to focus attention on clinically relevant differences, despite the several statistically significant differences across copayment groups. Diabetes therapy groups and the mean use of SMBG were similarly distributed across copayment groups.

Test Strip Utilization Following the Introduction of Free Test Strips

Parameter estimates for the 1-year change in the mean daily test strip utilization following the introduction of free test strips, after adjusting for secular trends observed in the reference group, were inconsistent with our study hypothesis (

). We had hypothesized that SMBG would increase after the provision of free test strips following the policy change and that the degree of this increase would be proportional to the amount of out-of-pocket cost paid per test strip prescription during the year before this policy change. In the OHA and MNT groups, there were no substantive or statistically significant increases in SMBG relative to the control group. There was no evidence of price elasticity in response to the change in cost in either group. In the insulin group, providing free test strips was accompanied by a counterintuitive decrease, albeit small (range, <0.5-2.5 fewer test strips per month), in SMBG among all copayment groups relative to the reference group.

Although the change in cost sharing (ie, each member's preexisting copayment amount) should not be affected by individual-level variables, we ran adjusted models to rule out potential confounding. The parameter estimates did not change substantively with adjustment, so we present the results from the more parsimonious (unadjusted) models. Furthermore, there was no evidence that socioeconomic status modified our findings, as there were no statistically significant differences in price elasticity among those who lived in predominantly impoverished vs nonimpoverished neighborhoods.

Test Strip Utilization Following the Introduction of a 20% Coinsurance

There were 15 646 subjects in the coinsurance cohort: 4505 subjects in the insulin group, 9627 subjects in the OHA group, and 1514 subjects in the MNT group. Fifty-eight percent of the cohort continued to receive test strips for free after the policy change, and 42% were charged the coinsur-ance ($8-$12). Subject characteristics did not differ clinically among those who were exposed vs those who were unexposed to the second policy change (a 20% coinsurance per prescription), although many of the contrasts were statistically significant because of the large sample sizes (

As in the previous analysis, we ran simple and adjusted models to determine the effect of this policy change on SMBG. Adjustment for individual-level covariates did not alter our results (ie, they were not confounders), so we report the results of models that adjust for prepolicy SMBG only. In the year following the policy change, subjects in the insulin, OHA, and MNT groups who were charged the coinsurance decreased their SMBG on average by 0.09 (

= .04) test strips per day, respectively, compared with those who continued to receive test strips for free. Although statistically significant, this decrease (approximately 2 test strips/month) is too small to be of clinical relevance.

In addition, there was no statistically significant difference in the SMBG discontinuation rates between those who were and those who were not charged the coinsurance. There was no statistically significant difference in the degree of the decline by socioeconomic status (ie, those living in vs those not living in impoverished neighborhoods) (data not shown).

This natural experiment provided a unique opportunity to evaluate the effect of changes in out-of-pocket costs on individual and population-level utilization of SMBG. We observed the effect of 2 policy-driven changes in cost sharing on SMBG (measured as test strip utilization) in a large managed care population with DM. As expected, we observed price-elastic demand in cross-sectional assessment of SMBG and out-of-pocket costs among prevalent SMBG users, suggesting the presence of a financial barrier to SMBG utilization. This graded utilization pattern was less marked when we restricted our analysis to the subset of patients with DM who were continuous SMBG users. This suggests that there may be a price threshold over which individuals choose to discontinue SMBG practice altogether, rather than to incrementally reduce test strip utilization (and SMBG frequency) in response to increasing test strip costs. Moreover, once a patient decides to initiate SMBG practice, he or she may be less price sensitive.

Our study findings are often counterintuitive and provide new insight into the effect of changes in healthcare policy, cost sharing, and provision of free services on this diabetes self-management practice. We expected that the elimination of the copayment would stimulate greater utilization, especially among those charged the most in the prepolicy period or among the poor. Paradoxically, the offering of free test strips following the implementation of SB64 did not increase SMBG above and beyond that expected by secular changes, despite historical evidence of price-elastic demand (based on the cross-sectional patterns). In fact, insulin users decreased test strip use slightly. Even economically disadvantaged patients and those paying the most for test strips before the policy change did not take extra advantage of the free test strip offer.

The SMBG practice patterns formed during the copay-ment period (before the implementation of SB64) persisted, despite receiving free test strips for 2 years. These entrenched patterns suggest that monitoring frequency may be established at the initiation of SMBG practice, affected by cost and other previously demonstrated predictors such as socioeconomic status, inconvenience, physical discomfort associated with the fingersticks, disease severity, perceived health benefits, provider recommendations, and family support.

However, once individuals have habituated to a stable SMBG practice, it is likely to be less quickly altered by access to free test strips or by even slight increases in cost. It is important to note that the health economic concern that providing free test strips would lead to a spike in excessive utilization (eg, multiple daily tests for oral agent only—treated patients) was unfounded; the prevalence of excessive use of SMBG continued to be dwarfed by the prevalence of underuse (<5% overutilized vs 40% did not practice at all in 2002 based on American Diabetes Association

clinical recommendations) after providing free test strips. Although the introduction of a small 20% coinsurance ($8-$12), the second policy change in the same population, resulted in a statistically significant decline in SMBG, it was too slight to be considered clinically relevant, and the decline did not differ by socioeconomic status.

Numerous cross-sectional studies have reported less frequent SMBG practice among patients lacking health insurance

The association with out-of-pocket costs (cost sharing) is also evident in studies

of fully insured managed care populations. Utilization of SMBG was shown to be more sensitive to out-of-pocket costs than other diabetes preventive services,

likely because demand is more price elastic when patients perceive a health service as optional.

There have been a few quasiexperimental or experimental studies. In one New England state, mandated policy change resulted in provision of free glucose meters, which was associated with increased SMBG use and with improved glycemic control,

although adherence to SMBG practice was of shorter duration among African Americans.

of a policy change by Veterans Affairs guidelines that reduced the number of test strips provided free of charge reported an associated reduction in SMBG among oral agent—treated patients with DM but no change in glycemic control. Two randomized studies of Canadian reimbursement policies demonstrated that providing free test strips was associated with increased SMBG among insulin-treated patients

but showed a significant improvement in glycemic control only in the former. Cost savings attributable to the receipt of free test strips differ substantially across studies, rendering the findings difficult to compare.

In a review of health insurance and healthcare utilization, Selby and Swain

noted that financial barriers negatively affect access among the poor. However, the evidence supporting higher price sensitivity for SMBG among the poor is inconsistent. A cross-sectional study

of SMBG in the KPNC population found significantly more price sensitivity among patients living in economically disadvantaged neighborhoods. However, the Translating Research Into Action for Diabetes study

of managed care across the United States failed to detect a significant interaction between income and copay-ment for test strips. The present study also failed to detect a differential response to increased or decreased cost sharing among subjects with varying financial advantage; therefore, provision of free test strips did not reduce social disparities in SMBG utilization.

We hypothesized a "symmetric" price-elastic demand for test strips and predicted that the copayment elimination would lead to increased SMBG and that introduction of a similar-sized coinsurance would lead to a similar-sized decrease in practice. Instead, we observed no response to the introduction of free test strips but observed a statistically significant, albeit not clinically relevant, response to the introduction of coinsurance. Asymmetry in price-elastic demand may be explained by behavioral economic theory. Rabin

has described the behavioral economic theory of loss aversion as follows: "The displeasure from a monetary loss is greater than the pleasure from a same-sized gain."

The elimination of the copayment (free test strips) and the subsequent out-of-pocket expense associated with the coinsurance may be viewed as a gain and as a subsequent loss. Statistically (albeit not clinically), our findings support Rabin's theory, which would predict a greater response to the loss.

There are limitations of our study that need to be noted. This study relied on the number of dispensed test strips as a proxy for SMBG practice; while we can never know how test strips dispensed are actually used, this metric has been validated previously.

Also, because members were not notified of policy changes, we restricted the study of change to ongoing SMBG users who would have been exposed to the cost changes at the pharmacy checkout. Therefore, we could not study nonusers at baseline, who might have initiated test strip use when test strips became free. It is possible that the lack of increase in SMBG among prevalent users once test strips became free was due to a lack of awareness of the new policy. However, this is unlikely, because patients responded to the converse situation; they decreased SMBG when prices increased. The range in out-of-pocket charges for test strips was modest, and we cannot predict responses to greater policy-related cost changes. Our secondary analysis on the effect of free test strips on social disparities may have been somewhat limited by our exclusion of poor patients with federally subsidized benefits. However, we have a substantial number of patients with regular (ie, commercial) benefits who would be considered financially disadvantaged, so we should be able to address this issue. The true cost of test strips must include all costs, such as the discomfort of testing and inconvenience; the elimination of the copayment does not diminish the discomfort of testing, and the economic cost may be trivial relative to these other unmeasured costs. The practice of SMBG and the sensitivity to cost in this insured population might be different from those observed in the general population. However, approximately one third of northern California's population receives its care from KPNC, and its membership's sociodemographic and self-reported health characteristics are consistent with those of the population in the surrounding geographic region.

This study has several strengths worth noting. The longitudinal analysis of this natural experiment with control subjects who are unaffected by the policy change provides more reliable evidence regarding the effect of financial barriers (or their removal) than cross-sectional studies. Study findings are especially compelling because the exposure to policy change and to cost sharing was unrelated to individual characteristics (ie, exogenous), as it is applied uniformly to groups of members whose benefit coverage is determined by their employer group contract, and individuals cannot ordinarily choose their cost-sharing level. Given that most persons in the United States with DM have health insurance (approximately 93% of patients with DM

), the findings should be relevant to other insured populations. In addition, the uniform access to care in KPNC provides a unique setting to study the practice of SMBG relative to benefit structures, in contrast to population-based studies in which healthcare access and quality vary. Similar to the randomized Health Insurance Experiment,

we found less price elasticity than many cross-sectional investigations, likely because of uncontrolled adverse selection in the cross-sectional study designs and to timescale issues outlined herein.

Given the need to control healthcare costs and to promote efforts to practice evidence-based medicine, SMBG guidelines and cost sharing are hotly debated in many managed care organizations and in countries with socialized medicine. Unexpectedly, providing free test strips did not lead to an unwanted spike in excessive utilization or even stimulate more guideline-adherent SMBG practice. This generous benefit cost the health plan in terms of lost cost-sharing revenues, without offset by measurable member health benefits or known downstream savings. The introduction of coinsur-ance shifted some of the economic burden back to the patient and slightly reduced demand (albeit unlikely causing clinical harm) in a population already underusing SMBG according to current clinical recommendations.

Utilization predictions are not always intuitive; we found entrenched copayment utilization patterns that persisted into the free test strip period and evidence suggestive of asymmetric price-elastic demand and loss aversion. In conclusion, cross-sectional relationships are likely not useful in forecasting utilization in the face of rapidly evolving benefit policies.

Convincing patients to adequately self-manage their DM is a daunting public health challenge. Cost sharing is implemented to minimize excessive utilization but is a blunt instrument, unintentionally also hindering recommended utilization. Because most patients underutilize and rarely overutilize SMBG (based on current clinical guidelines) even when receiving free test strips, the usefulness of creating additional financial barriers through cost sharing needs to be questioned. Future increases in cost sharing for medications may impose additional financial barriers that will further tip the balance, and patients may react more strongly to subsequent increases in out-of-pocket costs for SMBG. However, SMBG test strips pose a substantial cost to healthcare systems; therefore, future clinical guidelines need to carefully weigh the value of this practice against competing health expenditures.

: From the Division of Research (AJK, MMP, HHM, ATA, JVS) and Pharmacy Outcomes Research Group (JC, MMS), Kaiser Permanente, Oakland, Calif; and Department of Medicine, The David Geffen School of Medicine at UCLA, University of California, Los Angeles (SLE).

: This study was supported by grant R01 DK61678-02 from the National Institute of Diabetes and Digestive and Kidney Diseases. The funder had no role in the design or conduct of the study; collection, management, analysis, or interpretation of the data; or preparation, review, or approval of the manuscript. Dr Karter had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

: Andrew J. Karter, PhD, Division of Research, Kaiser Permanente, 2000 Broadway, Oakland, CA 94612. E-mail: andy.j.karter@kp.org.

: The authors (AJK, JC, HHM, MMS) report no relationship or financial interest with any entity that would pose a conflict of interest with the subject matter discussed in this manuscript. SLE reports receiving honorarium from Eli Lilly to review articles on cost-effectiveness of antipsychotic drugs. 

: Concept and design (AJK, HHM, MMS), acquisition of data (AJK, MMS) analysis and interpretation of data (JC, AJK, HHM, MMS, SLE), drafting of the manuscript (AJK, JC, HHM, SLE), and critical revision of the manuscript for important intellectual content (AJK, HH, MMS, SLE); acquisition of data (AJK, MMS); statistical analysis (AJK, SLE); obtaining funding (AJK) and administrative, technical, or logical support (AJK, HHM); supervision (AJK).

1. Diabetes Control and Complications Trial Research Group.

The effect of intensive treatment of diabetes on the development and progression of long-term complications in insulin-dependent diabetes melli-tus. 

2. Stratton IM, Adler AI, Neil HA, et al. 

Association of glycaemia with macrovascular and microvascular complications of type 2 diabetes (UKPDS 35): prospective observational study. 

3. Karter AJ, Ferrara A, Darbinian J, Ackerson LM, Selby JV. 

Self-monitoring of blood glucose: language and financial barriers in a managed care population with diabetes. 

4. Karter AJ, Stevens MR, Herman WH, et al. 

Out-of-pocket costs and diabetes preventive services: the Translating Research Into Action for Diabetes (TRIAD) study. 

Ameena T. Ahmed, MD, MPH

Articles

Mental Health Services Utilization Among Women Experiencing Intimate Partner Violence

Effect of Cost-sharing Changes on Self-monitoring of Blood Glucose